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 <asm/uaccess.h>
44 /* [Feb-1997 T. Schoebel-Theuer]
45 * Fundamental changes in the pathname lookup mechanisms (namei)
46 * were necessary because of omirr. The reason is that omirr needs
47 * to know the _real_ pathname, not the user-supplied one, in case
48 * of symlinks (and also when transname replacements occur).
50 * The new code replaces the old recursive symlink resolution with
51 * an iterative one (in case of non-nested symlink chains). It does
52 * this with calls to <fs>_follow_link().
53 * As a side effect, dir_namei(), _namei() and follow_link() are now
54 * replaced with a single function lookup_dentry() that can handle all
55 * the special cases of the former code.
57 * With the new dcache, the pathname is stored at each inode, at least as
58 * long as the refcount of the inode is positive. As a side effect, the
59 * size of the dcache depends on the inode cache and thus is dynamic.
61 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
62 * resolution to correspond with current state of the code.
64 * Note that the symlink resolution is not *completely* iterative.
65 * There is still a significant amount of tail- and mid- recursion in
66 * the algorithm. Also, note that <fs>_readlink() is not used in
67 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
68 * may return different results than <fs>_follow_link(). Many virtual
69 * filesystems (including /proc) exhibit this behavior.
72 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
73 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
74 * and the name already exists in form of a symlink, try to create the new
75 * name indicated by the symlink. The old code always complained that the
76 * name already exists, due to not following the symlink even if its target
77 * is nonexistent. The new semantics affects also mknod() and link() when
78 * the name is a symlink pointing to a non-existent name.
80 * I don't know which semantics is the right one, since I have no access
81 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
82 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
83 * "old" one. Personally, I think the new semantics is much more logical.
84 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
85 * file does succeed in both HP-UX and SunOs, but not in Solaris
86 * and in the old Linux semantics.
89 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
90 * semantics. See the comments in "open_namei" and "do_link" below.
92 * [10-Sep-98 Alan Modra] Another symlink change.
95 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
96 * inside the path - always follow.
97 * in the last component in creation/removal/renaming - never follow.
98 * if LOOKUP_FOLLOW passed - follow.
99 * if the pathname has trailing slashes - follow.
100 * otherwise - don't follow.
101 * (applied in that order).
103 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
104 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
105 * During the 2.4 we need to fix the userland stuff depending on it -
106 * hopefully we will be able to get rid of that wart in 2.5. So far only
107 * XEmacs seems to be relying on it...
110 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
111 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
112 * any extra contention...
115 /* In order to reduce some races, while at the same time doing additional
116 * checking and hopefully speeding things up, we copy filenames to the
117 * kernel data space before using them..
119 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
120 * PATH_MAX includes the nul terminator --RR.
123 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
126 getname_flags(const char __user *filename, int flags, int *empty)
128 struct filename *result;
132 result = audit_reusename(filename);
136 result = __getname();
137 if (unlikely(!result))
138 return ERR_PTR(-ENOMEM);
141 * First, try to embed the struct filename inside the names_cache
144 kname = (char *)result->iname;
145 result->name = kname;
147 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
148 if (unlikely(len < 0)) {
154 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
155 * separate struct filename so we can dedicate the entire
156 * names_cache allocation for the pathname, and re-do the copy from
159 if (unlikely(len == EMBEDDED_NAME_MAX)) {
160 const size_t size = offsetof(struct filename, iname[1]);
161 kname = (char *)result;
164 * size is chosen that way we to guarantee that
165 * result->iname[0] is within the same object and that
166 * kname can't be equal to result->iname, no matter what.
168 result = kzalloc(size, GFP_KERNEL);
169 if (unlikely(!result)) {
171 return ERR_PTR(-ENOMEM);
173 result->name = kname;
174 len = strncpy_from_user(kname, filename, PATH_MAX);
175 if (unlikely(len < 0)) {
180 if (unlikely(len == PATH_MAX)) {
183 return ERR_PTR(-ENAMETOOLONG);
188 /* The empty path is special. */
189 if (unlikely(!len)) {
192 if (!(flags & LOOKUP_EMPTY)) {
194 return ERR_PTR(-ENOENT);
198 result->uptr = filename;
199 result->aname = NULL;
200 audit_getname(result);
205 getname(const char __user * filename)
207 return getname_flags(filename, 0, NULL);
211 getname_kernel(const char * filename)
213 struct filename *result;
214 int len = strlen(filename) + 1;
216 result = __getname();
217 if (unlikely(!result))
218 return ERR_PTR(-ENOMEM);
220 if (len <= EMBEDDED_NAME_MAX) {
221 result->name = (char *)result->iname;
222 } else if (len <= PATH_MAX) {
223 struct filename *tmp;
225 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
226 if (unlikely(!tmp)) {
228 return ERR_PTR(-ENOMEM);
230 tmp->name = (char *)result;
234 return ERR_PTR(-ENAMETOOLONG);
236 memcpy((char *)result->name, filename, len);
238 result->aname = NULL;
240 audit_getname(result);
245 void putname(struct filename *name)
247 BUG_ON(name->refcnt <= 0);
249 if (--name->refcnt > 0)
252 if (name->name != name->iname) {
253 __putname(name->name);
259 static int check_acl(struct inode *inode, int mask)
261 #ifdef CONFIG_FS_POSIX_ACL
262 struct posix_acl *acl;
264 if (mask & MAY_NOT_BLOCK) {
265 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
268 /* no ->get_acl() calls in RCU mode... */
269 if (is_uncached_acl(acl))
271 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
274 acl = get_acl(inode, ACL_TYPE_ACCESS);
278 int error = posix_acl_permission(inode, acl, mask);
279 posix_acl_release(acl);
288 * This does the basic permission checking
290 static int acl_permission_check(struct inode *inode, int mask)
292 unsigned int mode = inode->i_mode;
294 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
297 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
298 int error = check_acl(inode, mask);
299 if (error != -EAGAIN)
303 if (in_group_p(inode->i_gid))
308 * If the DACs are ok we don't need any capability check.
310 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
316 * generic_permission - check for access rights on a Posix-like filesystem
317 * @inode: inode to check access rights for
318 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320 * Used to check for read/write/execute permissions on a file.
321 * We use "fsuid" for this, letting us set arbitrary permissions
322 * for filesystem access without changing the "normal" uids which
323 * are used for other things.
325 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326 * request cannot be satisfied (eg. requires blocking or too much complexity).
327 * It would then be called again in ref-walk mode.
329 int generic_permission(struct inode *inode, int mask)
334 * Do the basic permission checks.
336 ret = acl_permission_check(inode, mask);
340 if (S_ISDIR(inode->i_mode)) {
341 /* DACs are overridable for directories */
342 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344 if (!(mask & MAY_WRITE))
345 if (capable_wrt_inode_uidgid(inode,
346 CAP_DAC_READ_SEARCH))
351 * Read/write DACs are always overridable.
352 * Executable DACs are overridable when there is
353 * at least one exec bit set.
355 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
356 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
360 * Searching includes executable on directories, else just read.
362 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
363 if (mask == MAY_READ)
364 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
369 EXPORT_SYMBOL(generic_permission);
372 * We _really_ want to just do "generic_permission()" without
373 * even looking at the inode->i_op values. So we keep a cache
374 * flag in inode->i_opflags, that says "this has not special
375 * permission function, use the fast case".
377 static inline int do_inode_permission(struct inode *inode, int mask)
379 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
380 if (likely(inode->i_op->permission))
381 return inode->i_op->permission(inode, mask);
383 /* This gets set once for the inode lifetime */
384 spin_lock(&inode->i_lock);
385 inode->i_opflags |= IOP_FASTPERM;
386 spin_unlock(&inode->i_lock);
388 return generic_permission(inode, mask);
392 * __inode_permission - Check for access rights to a given inode
393 * @inode: Inode to check permission on
394 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396 * Check for read/write/execute permissions on an inode.
398 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400 * This does not check for a read-only file system. You probably want
401 * inode_permission().
403 int __inode_permission(struct inode *inode, int mask)
407 if (unlikely(mask & MAY_WRITE)) {
409 * Nobody gets write access to an immutable file.
411 if (IS_IMMUTABLE(inode))
415 retval = do_inode_permission(inode, mask);
419 retval = devcgroup_inode_permission(inode, mask);
423 return security_inode_permission(inode, mask);
425 EXPORT_SYMBOL(__inode_permission);
428 * sb_permission - Check superblock-level permissions
429 * @sb: Superblock of inode to check permission on
430 * @inode: Inode to check permission on
431 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
433 * Separate out file-system wide checks from inode-specific permission checks.
435 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
437 if (unlikely(mask & MAY_WRITE)) {
438 umode_t mode = inode->i_mode;
440 /* Nobody gets write access to a read-only fs. */
441 if ((sb->s_flags & MS_RDONLY) &&
442 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
449 * inode_permission - Check for access rights to a given inode
450 * @inode: Inode to check permission on
451 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
453 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
454 * this, letting us set arbitrary permissions for filesystem access without
455 * changing the "normal" UIDs which are used for other things.
457 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
459 int inode_permission(struct inode *inode, int mask)
463 retval = sb_permission(inode->i_sb, inode, mask);
466 return __inode_permission(inode, mask);
468 EXPORT_SYMBOL(inode_permission);
471 * path_get - get a reference to a path
472 * @path: path to get the reference to
474 * Given a path increment the reference count to the dentry and the vfsmount.
476 void path_get(const struct path *path)
481 EXPORT_SYMBOL(path_get);
484 * path_put - put a reference to a path
485 * @path: path to put the reference to
487 * Given a path decrement the reference count to the dentry and the vfsmount.
489 void path_put(const struct path *path)
494 EXPORT_SYMBOL(path_put);
496 #define EMBEDDED_LEVELS 2
501 struct inode *inode; /* path.dentry.d_inode */
506 int total_link_count;
509 struct delayed_call done;
512 } *stack, internal[EMBEDDED_LEVELS];
513 struct filename *name;
514 struct nameidata *saved;
515 struct inode *link_inode;
520 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
522 struct nameidata *old = current->nameidata;
523 p->stack = p->internal;
526 p->total_link_count = old ? old->total_link_count : 0;
528 current->nameidata = p;
531 static void restore_nameidata(void)
533 struct nameidata *now = current->nameidata, *old = now->saved;
535 current->nameidata = old;
537 old->total_link_count = now->total_link_count;
538 if (now->stack != now->internal)
542 static int __nd_alloc_stack(struct nameidata *nd)
546 if (nd->flags & LOOKUP_RCU) {
547 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
552 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557 memcpy(p, nd->internal, sizeof(nd->internal));
563 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
564 * @path: nameidate to verify
566 * Rename can sometimes move a file or directory outside of a bind
567 * mount, path_connected allows those cases to be detected.
569 static bool path_connected(const struct path *path)
571 struct vfsmount *mnt = path->mnt;
573 /* Only bind mounts can have disconnected paths */
574 if (mnt->mnt_root == mnt->mnt_sb->s_root)
577 return is_subdir(path->dentry, mnt->mnt_root);
580 static inline int nd_alloc_stack(struct nameidata *nd)
582 if (likely(nd->depth != EMBEDDED_LEVELS))
584 if (likely(nd->stack != nd->internal))
586 return __nd_alloc_stack(nd);
589 static void drop_links(struct nameidata *nd)
593 struct saved *last = nd->stack + i;
594 do_delayed_call(&last->done);
595 clear_delayed_call(&last->done);
599 static void terminate_walk(struct nameidata *nd)
602 if (!(nd->flags & LOOKUP_RCU)) {
605 for (i = 0; i < nd->depth; i++)
606 path_put(&nd->stack[i].link);
607 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
612 nd->flags &= ~LOOKUP_RCU;
613 if (!(nd->flags & LOOKUP_ROOT))
620 /* path_put is needed afterwards regardless of success or failure */
621 static bool legitimize_path(struct nameidata *nd,
622 struct path *path, unsigned seq)
624 int res = __legitimize_mnt(path->mnt, nd->m_seq);
631 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
635 return !read_seqcount_retry(&path->dentry->d_seq, seq);
638 static bool legitimize_links(struct nameidata *nd)
641 for (i = 0; i < nd->depth; i++) {
642 struct saved *last = nd->stack + i;
643 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 * Path walking has 2 modes, rcu-walk and ref-walk (see
654 * Documentation/filesystems/path-lookup.txt). In situations when we can't
655 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
656 * normal reference counts on dentries and vfsmounts to transition to ref-walk
657 * mode. Refcounts are grabbed at the last known good point before rcu-walk
658 * got stuck, so ref-walk may continue from there. If this is not successful
659 * (eg. a seqcount has changed), then failure is returned and it's up to caller
660 * to restart the path walk from the beginning in ref-walk mode.
664 * unlazy_walk - try to switch to ref-walk mode.
665 * @nd: nameidata pathwalk data
666 * @dentry: child of nd->path.dentry or NULL
667 * @seq: seq number to check dentry against
668 * Returns: 0 on success, -ECHILD on failure
670 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
671 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
672 * @nd or NULL. Must be called from rcu-walk context.
673 * Nothing should touch nameidata between unlazy_walk() failure and
676 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
678 struct dentry *parent = nd->path.dentry;
680 BUG_ON(!(nd->flags & LOOKUP_RCU));
682 nd->flags &= ~LOOKUP_RCU;
683 if (unlikely(!legitimize_links(nd)))
685 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
687 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
691 * For a negative lookup, the lookup sequence point is the parents
692 * sequence point, and it only needs to revalidate the parent dentry.
694 * For a positive lookup, we need to move both the parent and the
695 * dentry from the RCU domain to be properly refcounted. And the
696 * sequence number in the dentry validates *both* dentry counters,
697 * since we checked the sequence number of the parent after we got
698 * the child sequence number. So we know the parent must still
699 * be valid if the child sequence number is still valid.
702 if (read_seqcount_retry(&parent->d_seq, nd->seq))
704 BUG_ON(nd->inode != parent->d_inode);
706 if (!lockref_get_not_dead(&dentry->d_lockref))
708 if (read_seqcount_retry(&dentry->d_seq, seq))
713 * Sequence counts matched. Now make sure that the root is
714 * still valid and get it if required.
716 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
717 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
734 nd->path.dentry = NULL;
738 if (!(nd->flags & LOOKUP_ROOT))
743 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
745 if (unlikely(!legitimize_path(nd, link, seq))) {
748 nd->flags &= ~LOOKUP_RCU;
750 nd->path.dentry = NULL;
751 if (!(nd->flags & LOOKUP_ROOT))
754 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
761 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
763 return dentry->d_op->d_revalidate(dentry, flags);
767 * complete_walk - successful completion of path walk
768 * @nd: pointer nameidata
770 * If we had been in RCU mode, drop out of it and legitimize nd->path.
771 * Revalidate the final result, unless we'd already done that during
772 * the path walk or the filesystem doesn't ask for it. Return 0 on
773 * success, -error on failure. In case of failure caller does not
774 * need to drop nd->path.
776 static int complete_walk(struct nameidata *nd)
778 struct dentry *dentry = nd->path.dentry;
781 if (nd->flags & LOOKUP_RCU) {
782 if (!(nd->flags & LOOKUP_ROOT))
784 if (unlikely(unlazy_walk(nd, NULL, 0)))
788 if (likely(!(nd->flags & LOOKUP_JUMPED)))
791 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
794 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
804 static void set_root(struct nameidata *nd)
806 struct fs_struct *fs = current->fs;
808 if (nd->flags & LOOKUP_RCU) {
812 seq = read_seqcount_begin(&fs->seq);
814 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
815 } while (read_seqcount_retry(&fs->seq, seq));
817 get_fs_root(fs, &nd->root);
821 static void path_put_conditional(struct path *path, struct nameidata *nd)
824 if (path->mnt != nd->path.mnt)
828 static inline void path_to_nameidata(const struct path *path,
829 struct nameidata *nd)
831 if (!(nd->flags & LOOKUP_RCU)) {
832 dput(nd->path.dentry);
833 if (nd->path.mnt != path->mnt)
834 mntput(nd->path.mnt);
836 nd->path.mnt = path->mnt;
837 nd->path.dentry = path->dentry;
840 static int nd_jump_root(struct nameidata *nd)
842 if (nd->flags & LOOKUP_RCU) {
846 nd->inode = d->d_inode;
847 nd->seq = nd->root_seq;
848 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
854 nd->inode = nd->path.dentry->d_inode;
856 nd->flags |= LOOKUP_JUMPED;
861 * Helper to directly jump to a known parsed path from ->get_link,
862 * caller must have taken a reference to path beforehand.
864 void nd_jump_link(struct path *path)
866 struct nameidata *nd = current->nameidata;
870 nd->inode = nd->path.dentry->d_inode;
871 nd->flags |= LOOKUP_JUMPED;
874 static inline void put_link(struct nameidata *nd)
876 struct saved *last = nd->stack + --nd->depth;
877 do_delayed_call(&last->done);
878 if (!(nd->flags & LOOKUP_RCU))
879 path_put(&last->link);
882 int sysctl_protected_symlinks __read_mostly = 0;
883 int sysctl_protected_hardlinks __read_mostly = 0;
886 * may_follow_link - Check symlink following for unsafe situations
887 * @nd: nameidata pathwalk data
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
898 * Returns 0 if following the symlink is allowed, -ve on error.
900 static inline int may_follow_link(struct nameidata *nd)
902 const struct inode *inode;
903 const struct inode *parent;
905 if (!sysctl_protected_symlinks)
908 /* Allowed if owner and follower match. */
909 inode = nd->link_inode;
910 if (uid_eq(current_cred()->fsuid, inode->i_uid))
913 /* Allowed if parent directory not sticky and world-writable. */
915 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
918 /* Allowed if parent directory and link owner match. */
919 if (uid_eq(parent->i_uid, inode->i_uid))
922 if (nd->flags & LOOKUP_RCU)
925 audit_log_link_denied("follow_link", &nd->stack[0].link);
930 * safe_hardlink_source - Check for safe hardlink conditions
931 * @inode: the source inode to hardlink from
933 * Return false if at least one of the following conditions:
934 * - inode is not a regular file
936 * - inode is setgid and group-exec
937 * - access failure for read and write
939 * Otherwise returns true.
941 static bool safe_hardlink_source(struct inode *inode)
943 umode_t mode = inode->i_mode;
945 /* Special files should not get pinned to the filesystem. */
949 /* Setuid files should not get pinned to the filesystem. */
953 /* Executable setgid files should not get pinned to the filesystem. */
954 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
957 /* Hardlinking to unreadable or unwritable sources is dangerous. */
958 if (inode_permission(inode, MAY_READ | MAY_WRITE))
965 * may_linkat - Check permissions for creating a hardlink
966 * @link: the source to hardlink from
968 * Block hardlink when all of:
969 * - sysctl_protected_hardlinks enabled
970 * - fsuid does not match inode
971 * - hardlink source is unsafe (see safe_hardlink_source() above)
972 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
974 * Returns 0 if successful, -ve on error.
976 static int may_linkat(struct path *link)
980 if (!sysctl_protected_hardlinks)
983 inode = link->dentry->d_inode;
985 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
986 * otherwise, it must be a safe source.
988 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
991 audit_log_link_denied("linkat", link);
995 static __always_inline
996 const char *get_link(struct nameidata *nd)
998 struct saved *last = nd->stack + nd->depth - 1;
999 struct dentry *dentry = last->link.dentry;
1000 struct inode *inode = nd->link_inode;
1004 if (!(nd->flags & LOOKUP_RCU)) {
1005 touch_atime(&last->link);
1007 } else if (atime_needs_update(&last->link, inode)) {
1008 if (unlikely(unlazy_walk(nd, NULL, 0)))
1009 return ERR_PTR(-ECHILD);
1010 touch_atime(&last->link);
1013 error = security_inode_follow_link(dentry, inode,
1014 nd->flags & LOOKUP_RCU);
1015 if (unlikely(error))
1016 return ERR_PTR(error);
1018 nd->last_type = LAST_BIND;
1019 res = inode->i_link;
1021 const char * (*get)(struct dentry *, struct inode *,
1022 struct delayed_call *);
1023 get = inode->i_op->get_link;
1024 if (nd->flags & LOOKUP_RCU) {
1025 res = get(NULL, inode, &last->done);
1026 if (res == ERR_PTR(-ECHILD)) {
1027 if (unlikely(unlazy_walk(nd, NULL, 0)))
1028 return ERR_PTR(-ECHILD);
1029 res = get(dentry, inode, &last->done);
1032 res = get(dentry, inode, &last->done);
1034 if (IS_ERR_OR_NULL(res))
1040 if (unlikely(nd_jump_root(nd)))
1041 return ERR_PTR(-ECHILD);
1042 while (unlikely(*++res == '/'))
1051 * follow_up - Find the mountpoint of path's vfsmount
1053 * Given a path, find the mountpoint of its source file system.
1054 * Replace @path with the path of the mountpoint in the parent mount.
1057 * Return 1 if we went up a level and 0 if we were already at the
1060 int follow_up(struct path *path)
1062 struct mount *mnt = real_mount(path->mnt);
1063 struct mount *parent;
1064 struct dentry *mountpoint;
1066 read_seqlock_excl(&mount_lock);
1067 parent = mnt->mnt_parent;
1068 if (parent == mnt) {
1069 read_sequnlock_excl(&mount_lock);
1072 mntget(&parent->mnt);
1073 mountpoint = dget(mnt->mnt_mountpoint);
1074 read_sequnlock_excl(&mount_lock);
1076 path->dentry = mountpoint;
1078 path->mnt = &parent->mnt;
1081 EXPORT_SYMBOL(follow_up);
1084 * Perform an automount
1085 * - return -EISDIR to tell follow_managed() to stop and return the path we
1088 static int follow_automount(struct path *path, struct nameidata *nd,
1091 struct vfsmount *mnt;
1094 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1097 /* We don't want to mount if someone's just doing a stat -
1098 * unless they're stat'ing a directory and appended a '/' to
1101 * We do, however, want to mount if someone wants to open or
1102 * create a file of any type under the mountpoint, wants to
1103 * traverse through the mountpoint or wants to open the
1104 * mounted directory. Also, autofs may mark negative dentries
1105 * as being automount points. These will need the attentions
1106 * of the daemon to instantiate them before they can be used.
1108 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1109 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1110 path->dentry->d_inode)
1113 nd->total_link_count++;
1114 if (nd->total_link_count >= 40)
1117 mnt = path->dentry->d_op->d_automount(path);
1120 * The filesystem is allowed to return -EISDIR here to indicate
1121 * it doesn't want to automount. For instance, autofs would do
1122 * this so that its userspace daemon can mount on this dentry.
1124 * However, we can only permit this if it's a terminal point in
1125 * the path being looked up; if it wasn't then the remainder of
1126 * the path is inaccessible and we should say so.
1128 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1130 return PTR_ERR(mnt);
1133 if (!mnt) /* mount collision */
1136 if (!*need_mntput) {
1137 /* lock_mount() may release path->mnt on error */
1139 *need_mntput = true;
1141 err = finish_automount(mnt, path);
1145 /* Someone else made a mount here whilst we were busy */
1150 path->dentry = dget(mnt->mnt_root);
1159 * Handle a dentry that is managed in some way.
1160 * - Flagged for transit management (autofs)
1161 * - Flagged as mountpoint
1162 * - Flagged as automount point
1164 * This may only be called in refwalk mode.
1166 * Serialization is taken care of in namespace.c
1168 static int follow_managed(struct path *path, struct nameidata *nd)
1170 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1172 bool need_mntput = false;
1175 /* Given that we're not holding a lock here, we retain the value in a
1176 * local variable for each dentry as we look at it so that we don't see
1177 * the components of that value change under us */
1178 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1179 managed &= DCACHE_MANAGED_DENTRY,
1180 unlikely(managed != 0)) {
1181 /* Allow the filesystem to manage the transit without i_mutex
1183 if (managed & DCACHE_MANAGE_TRANSIT) {
1184 BUG_ON(!path->dentry->d_op);
1185 BUG_ON(!path->dentry->d_op->d_manage);
1186 ret = path->dentry->d_op->d_manage(path->dentry, false);
1191 /* Transit to a mounted filesystem. */
1192 if (managed & DCACHE_MOUNTED) {
1193 struct vfsmount *mounted = lookup_mnt(path);
1198 path->mnt = mounted;
1199 path->dentry = dget(mounted->mnt_root);
1204 /* Something is mounted on this dentry in another
1205 * namespace and/or whatever was mounted there in this
1206 * namespace got unmounted before lookup_mnt() could
1210 /* Handle an automount point */
1211 if (managed & DCACHE_NEED_AUTOMOUNT) {
1212 ret = follow_automount(path, nd, &need_mntput);
1218 /* We didn't change the current path point */
1222 if (need_mntput && path->mnt == mnt)
1224 if (ret == -EISDIR || !ret)
1227 nd->flags |= LOOKUP_JUMPED;
1228 if (unlikely(ret < 0))
1229 path_put_conditional(path, nd);
1233 int follow_down_one(struct path *path)
1235 struct vfsmount *mounted;
1237 mounted = lookup_mnt(path);
1241 path->mnt = mounted;
1242 path->dentry = dget(mounted->mnt_root);
1247 EXPORT_SYMBOL(follow_down_one);
1249 static inline int managed_dentry_rcu(struct dentry *dentry)
1251 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1252 dentry->d_op->d_manage(dentry, true) : 0;
1256 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1257 * we meet a managed dentry that would need blocking.
1259 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1260 struct inode **inode, unsigned *seqp)
1263 struct mount *mounted;
1265 * Don't forget we might have a non-mountpoint managed dentry
1266 * that wants to block transit.
1268 switch (managed_dentry_rcu(path->dentry)) {
1278 if (!d_mountpoint(path->dentry))
1279 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1281 mounted = __lookup_mnt(path->mnt, path->dentry);
1284 path->mnt = &mounted->mnt;
1285 path->dentry = mounted->mnt.mnt_root;
1286 nd->flags |= LOOKUP_JUMPED;
1287 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1289 * Update the inode too. We don't need to re-check the
1290 * dentry sequence number here after this d_inode read,
1291 * because a mount-point is always pinned.
1293 *inode = path->dentry->d_inode;
1295 return !read_seqretry(&mount_lock, nd->m_seq) &&
1296 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1299 static int follow_dotdot_rcu(struct nameidata *nd)
1301 struct inode *inode = nd->inode;
1304 if (path_equal(&nd->path, &nd->root))
1306 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307 struct dentry *old = nd->path.dentry;
1308 struct dentry *parent = old->d_parent;
1311 inode = parent->d_inode;
1312 seq = read_seqcount_begin(&parent->d_seq);
1313 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1315 nd->path.dentry = parent;
1317 if (unlikely(!path_connected(&nd->path)))
1321 struct mount *mnt = real_mount(nd->path.mnt);
1322 struct mount *mparent = mnt->mnt_parent;
1323 struct dentry *mountpoint = mnt->mnt_mountpoint;
1324 struct inode *inode2 = mountpoint->d_inode;
1325 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1326 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1328 if (&mparent->mnt == nd->path.mnt)
1330 /* we know that mountpoint was pinned */
1331 nd->path.dentry = mountpoint;
1332 nd->path.mnt = &mparent->mnt;
1337 while (unlikely(d_mountpoint(nd->path.dentry))) {
1338 struct mount *mounted;
1339 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1340 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1344 nd->path.mnt = &mounted->mnt;
1345 nd->path.dentry = mounted->mnt.mnt_root;
1346 inode = nd->path.dentry->d_inode;
1347 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1354 * Follow down to the covering mount currently visible to userspace. At each
1355 * point, the filesystem owning that dentry may be queried as to whether the
1356 * caller is permitted to proceed or not.
1358 int follow_down(struct path *path)
1363 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1364 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1365 /* Allow the filesystem to manage the transit without i_mutex
1368 * We indicate to the filesystem if someone is trying to mount
1369 * something here. This gives autofs the chance to deny anyone
1370 * other than its daemon the right to mount on its
1373 * The filesystem may sleep at this point.
1375 if (managed & DCACHE_MANAGE_TRANSIT) {
1376 BUG_ON(!path->dentry->d_op);
1377 BUG_ON(!path->dentry->d_op->d_manage);
1378 ret = path->dentry->d_op->d_manage(
1379 path->dentry, false);
1381 return ret == -EISDIR ? 0 : ret;
1384 /* Transit to a mounted filesystem. */
1385 if (managed & DCACHE_MOUNTED) {
1386 struct vfsmount *mounted = lookup_mnt(path);
1391 path->mnt = mounted;
1392 path->dentry = dget(mounted->mnt_root);
1396 /* Don't handle automount points here */
1401 EXPORT_SYMBOL(follow_down);
1404 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1406 static void follow_mount(struct path *path)
1408 while (d_mountpoint(path->dentry)) {
1409 struct vfsmount *mounted = lookup_mnt(path);
1414 path->mnt = mounted;
1415 path->dentry = dget(mounted->mnt_root);
1419 static int path_parent_directory(struct path *path)
1421 struct dentry *old = path->dentry;
1422 /* rare case of legitimate dget_parent()... */
1423 path->dentry = dget_parent(path->dentry);
1425 if (unlikely(!path_connected(path)))
1430 static int follow_dotdot(struct nameidata *nd)
1433 if (nd->path.dentry == nd->root.dentry &&
1434 nd->path.mnt == nd->root.mnt) {
1437 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1438 int ret = path_parent_directory(&nd->path);
1443 if (!follow_up(&nd->path))
1446 follow_mount(&nd->path);
1447 nd->inode = nd->path.dentry->d_inode;
1452 * This looks up the name in dcache, possibly revalidates the old dentry and
1453 * allocates a new one if not found or not valid. In the need_lookup argument
1454 * returns whether i_op->lookup is necessary.
1456 static struct dentry *lookup_dcache(const struct qstr *name,
1460 struct dentry *dentry;
1463 dentry = d_lookup(dir, name);
1465 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1466 error = d_revalidate(dentry, flags);
1467 if (unlikely(error <= 0)) {
1469 d_invalidate(dentry);
1471 return ERR_PTR(error);
1479 * Call i_op->lookup on the dentry. The dentry must be negative and
1482 * dir->d_inode->i_mutex must be held
1484 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1489 /* Don't create child dentry for a dead directory. */
1490 if (unlikely(IS_DEADDIR(dir))) {
1492 return ERR_PTR(-ENOENT);
1495 old = dir->i_op->lookup(dir, dentry, flags);
1496 if (unlikely(old)) {
1503 static struct dentry *__lookup_hash(const struct qstr *name,
1504 struct dentry *base, unsigned int flags)
1506 struct dentry *dentry = lookup_dcache(name, base, flags);
1511 dentry = d_alloc(base, name);
1512 if (unlikely(!dentry))
1513 return ERR_PTR(-ENOMEM);
1515 return lookup_real(base->d_inode, dentry, flags);
1518 static int lookup_fast(struct nameidata *nd,
1519 struct path *path, struct inode **inode,
1522 struct vfsmount *mnt = nd->path.mnt;
1523 struct dentry *dentry, *parent = nd->path.dentry;
1528 * Rename seqlock is not required here because in the off chance
1529 * of a false negative due to a concurrent rename, the caller is
1530 * going to fall back to non-racy lookup.
1532 if (nd->flags & LOOKUP_RCU) {
1535 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1536 if (unlikely(!dentry)) {
1537 if (unlazy_walk(nd, NULL, 0))
1543 * This sequence count validates that the inode matches
1544 * the dentry name information from lookup.
1546 *inode = d_backing_inode(dentry);
1547 negative = d_is_negative(dentry);
1548 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1552 * This sequence count validates that the parent had no
1553 * changes while we did the lookup of the dentry above.
1555 * The memory barrier in read_seqcount_begin of child is
1556 * enough, we can use __read_seqcount_retry here.
1558 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1562 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1563 status = d_revalidate(dentry, nd->flags);
1564 if (unlikely(status <= 0)) {
1565 if (unlazy_walk(nd, dentry, seq))
1567 if (status == -ECHILD)
1568 status = d_revalidate(dentry, nd->flags);
1571 * Note: do negative dentry check after revalidation in
1572 * case that drops it.
1574 if (unlikely(negative))
1577 path->dentry = dentry;
1578 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1580 if (unlazy_walk(nd, dentry, seq))
1584 dentry = __d_lookup(parent, &nd->last);
1585 if (unlikely(!dentry))
1587 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1588 status = d_revalidate(dentry, nd->flags);
1590 if (unlikely(status <= 0)) {
1592 d_invalidate(dentry);
1596 if (unlikely(d_is_negative(dentry))) {
1602 path->dentry = dentry;
1603 err = follow_managed(path, nd);
1604 if (likely(err > 0))
1605 *inode = d_backing_inode(path->dentry);
1609 /* Fast lookup failed, do it the slow way */
1610 static struct dentry *lookup_slow(const struct qstr *name,
1614 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1615 struct inode *inode = dir->d_inode;
1616 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1618 inode_lock_shared(inode);
1619 /* Don't go there if it's already dead */
1620 if (unlikely(IS_DEADDIR(inode)))
1623 dentry = d_alloc_parallel(dir, name, &wq);
1626 if (unlikely(!d_in_lookup(dentry))) {
1627 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1628 !(flags & LOOKUP_NO_REVAL)) {
1629 int error = d_revalidate(dentry, flags);
1630 if (unlikely(error <= 0)) {
1632 d_invalidate(dentry);
1637 dentry = ERR_PTR(error);
1641 old = inode->i_op->lookup(inode, dentry, flags);
1642 d_lookup_done(dentry);
1643 if (unlikely(old)) {
1649 inode_unlock_shared(inode);
1653 static inline int may_lookup(struct nameidata *nd)
1655 if (nd->flags & LOOKUP_RCU) {
1656 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1659 if (unlazy_walk(nd, NULL, 0))
1662 return inode_permission(nd->inode, MAY_EXEC);
1665 static inline int handle_dots(struct nameidata *nd, int type)
1667 if (type == LAST_DOTDOT) {
1670 if (nd->flags & LOOKUP_RCU) {
1671 return follow_dotdot_rcu(nd);
1673 return follow_dotdot(nd);
1678 static int pick_link(struct nameidata *nd, struct path *link,
1679 struct inode *inode, unsigned seq)
1683 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1684 path_to_nameidata(link, nd);
1687 if (!(nd->flags & LOOKUP_RCU)) {
1688 if (link->mnt == nd->path.mnt)
1691 error = nd_alloc_stack(nd);
1692 if (unlikely(error)) {
1693 if (error == -ECHILD) {
1694 if (unlikely(unlazy_link(nd, link, seq)))
1696 error = nd_alloc_stack(nd);
1704 last = nd->stack + nd->depth++;
1706 clear_delayed_call(&last->done);
1707 nd->link_inode = inode;
1713 * Do we need to follow links? We _really_ want to be able
1714 * to do this check without having to look at inode->i_op,
1715 * so we keep a cache of "no, this doesn't need follow_link"
1716 * for the common case.
1718 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1720 struct inode *inode, unsigned seq)
1722 if (likely(!d_is_symlink(link->dentry)))
1726 /* make sure that d_is_symlink above matches inode */
1727 if (nd->flags & LOOKUP_RCU) {
1728 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1731 return pick_link(nd, link, inode, seq);
1734 enum {WALK_GET = 1, WALK_PUT = 2};
1736 static int walk_component(struct nameidata *nd, int flags)
1739 struct inode *inode;
1743 * "." and ".." are special - ".." especially so because it has
1744 * to be able to know about the current root directory and
1745 * parent relationships.
1747 if (unlikely(nd->last_type != LAST_NORM)) {
1748 err = handle_dots(nd, nd->last_type);
1749 if (flags & WALK_PUT)
1753 err = lookup_fast(nd, &path, &inode, &seq);
1754 if (unlikely(err <= 0)) {
1757 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1759 if (IS_ERR(path.dentry))
1760 return PTR_ERR(path.dentry);
1762 path.mnt = nd->path.mnt;
1763 err = follow_managed(&path, nd);
1764 if (unlikely(err < 0))
1767 if (unlikely(d_is_negative(path.dentry))) {
1768 path_to_nameidata(&path, nd);
1772 seq = 0; /* we are already out of RCU mode */
1773 inode = d_backing_inode(path.dentry);
1776 if (flags & WALK_PUT)
1778 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1781 path_to_nameidata(&path, nd);
1788 * We can do the critical dentry name comparison and hashing
1789 * operations one word at a time, but we are limited to:
1791 * - Architectures with fast unaligned word accesses. We could
1792 * do a "get_unaligned()" if this helps and is sufficiently
1795 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1796 * do not trap on the (extremely unlikely) case of a page
1797 * crossing operation.
1799 * - Furthermore, we need an efficient 64-bit compile for the
1800 * 64-bit case in order to generate the "number of bytes in
1801 * the final mask". Again, that could be replaced with a
1802 * efficient population count instruction or similar.
1804 #ifdef CONFIG_DCACHE_WORD_ACCESS
1806 #include <asm/word-at-a-time.h>
1810 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1812 #elif defined(CONFIG_64BIT)
1814 * Register pressure in the mixing function is an issue, particularly
1815 * on 32-bit x86, but almost any function requires one state value and
1816 * one temporary. Instead, use a function designed for two state values
1817 * and no temporaries.
1819 * This function cannot create a collision in only two iterations, so
1820 * we have two iterations to achieve avalanche. In those two iterations,
1821 * we have six layers of mixing, which is enough to spread one bit's
1822 * influence out to 2^6 = 64 state bits.
1824 * Rotate constants are scored by considering either 64 one-bit input
1825 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1826 * probability of that delta causing a change to each of the 128 output
1827 * bits, using a sample of random initial states.
1829 * The Shannon entropy of the computed probabilities is then summed
1830 * to produce a score. Ideally, any input change has a 50% chance of
1831 * toggling any given output bit.
1833 * Mixing scores (in bits) for (12,45):
1834 * Input delta: 1-bit 2-bit
1835 * 1 round: 713.3 42542.6
1836 * 2 rounds: 2753.7 140389.8
1837 * 3 rounds: 5954.1 233458.2
1838 * 4 rounds: 7862.6 256672.2
1839 * Perfect: 8192 258048
1840 * (64*128) (64*63/2 * 128)
1842 #define HASH_MIX(x, y, a) \
1844 y ^= x, x = rol64(x,12),\
1845 x += y, y = rol64(y,45),\
1849 * Fold two longs into one 32-bit hash value. This must be fast, but
1850 * latency isn't quite as critical, as there is a fair bit of additional
1851 * work done before the hash value is used.
1853 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1855 y ^= x * GOLDEN_RATIO_64;
1856 y *= GOLDEN_RATIO_64;
1860 #else /* 32-bit case */
1863 * Mixing scores (in bits) for (7,20):
1864 * Input delta: 1-bit 2-bit
1865 * 1 round: 330.3 9201.6
1866 * 2 rounds: 1246.4 25475.4
1867 * 3 rounds: 1907.1 31295.1
1868 * 4 rounds: 2042.3 31718.6
1869 * Perfect: 2048 31744
1870 * (32*64) (32*31/2 * 64)
1872 #define HASH_MIX(x, y, a) \
1874 y ^= x, x = rol32(x, 7),\
1875 x += y, y = rol32(y,20),\
1878 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1880 /* Use arch-optimized multiply if one exists */
1881 return __hash_32(y ^ __hash_32(x));
1887 * Return the hash of a string of known length. This is carfully
1888 * designed to match hash_name(), which is the more critical function.
1889 * In particular, we must end by hashing a final word containing 0..7
1890 * payload bytes, to match the way that hash_name() iterates until it
1891 * finds the delimiter after the name.
1893 unsigned int full_name_hash(const char *name, unsigned int len)
1895 unsigned long a, x = 0, y = 0;
1900 a = load_unaligned_zeropad(name);
1901 if (len < sizeof(unsigned long))
1904 name += sizeof(unsigned long);
1905 len -= sizeof(unsigned long);
1907 x ^= a & bytemask_from_count(len);
1909 return fold_hash(x, y);
1911 EXPORT_SYMBOL(full_name_hash);
1913 /* Return the "hash_len" (hash and length) of a null-terminated string */
1914 u64 hashlen_string(const char *name)
1916 unsigned long a = 0, x = 0, y = 0, adata, mask, len;
1917 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1919 len = -sizeof(unsigned long);
1922 len += sizeof(unsigned long);
1923 a = load_unaligned_zeropad(name+len);
1924 } while (!has_zero(a, &adata, &constants));
1926 adata = prep_zero_mask(a, adata, &constants);
1927 mask = create_zero_mask(adata);
1928 x ^= a & zero_bytemask(mask);
1930 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1932 EXPORT_SYMBOL(hashlen_string);
1935 * Calculate the length and hash of the path component, and
1936 * return the "hash_len" as the result.
1938 static inline u64 hash_name(const char *name)
1940 unsigned long a = 0, b, x = 0, y = 0, adata, bdata, mask, len;
1941 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1943 len = -sizeof(unsigned long);
1946 len += sizeof(unsigned long);
1947 a = load_unaligned_zeropad(name+len);
1948 b = a ^ REPEAT_BYTE('/');
1949 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1951 adata = prep_zero_mask(a, adata, &constants);
1952 bdata = prep_zero_mask(b, bdata, &constants);
1953 mask = create_zero_mask(adata | bdata);
1954 x ^= a & zero_bytemask(mask);
1956 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1959 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1961 /* Return the hash of a string of known length */
1962 unsigned int full_name_hash(const char *name, unsigned int len)
1964 unsigned long hash = init_name_hash();
1966 hash = partial_name_hash((unsigned char)*name++, hash);
1967 return end_name_hash(hash);
1969 EXPORT_SYMBOL(full_name_hash);
1971 /* Return the "hash_len" (hash and length) of a null-terminated string */
1972 u64 hashlen_string(const char *name)
1974 unsigned long hash = init_name_hash();
1975 unsigned long len = 0, c;
1977 c = (unsigned char)*name;
1980 hash = partial_name_hash(c, hash);
1981 c = (unsigned char)name[len];
1983 return hashlen_create(end_name_hash(hash), len);
1985 EXPORT_SYMBOL(hashlen_string);
1988 * We know there's a real path component here of at least
1991 static inline u64 hash_name(const char *name)
1993 unsigned long hash = init_name_hash();
1994 unsigned long len = 0, c;
1996 c = (unsigned char)*name;
1999 hash = partial_name_hash(c, hash);
2000 c = (unsigned char)name[len];
2001 } while (c && c != '/');
2002 return hashlen_create(end_name_hash(hash), len);
2009 * This is the basic name resolution function, turning a pathname into
2010 * the final dentry. We expect 'base' to be positive and a directory.
2012 * Returns 0 and nd will have valid dentry and mnt on success.
2013 * Returns error and drops reference to input namei data on failure.
2015 static int link_path_walk(const char *name, struct nameidata *nd)
2024 /* At this point we know we have a real path component. */
2029 err = may_lookup(nd);
2033 hash_len = hash_name(name);
2036 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2038 if (name[1] == '.') {
2040 nd->flags |= LOOKUP_JUMPED;
2046 if (likely(type == LAST_NORM)) {
2047 struct dentry *parent = nd->path.dentry;
2048 nd->flags &= ~LOOKUP_JUMPED;
2049 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2050 struct qstr this = { { .hash_len = hash_len }, .name = name };
2051 err = parent->d_op->d_hash(parent, &this);
2054 hash_len = this.hash_len;
2059 nd->last.hash_len = hash_len;
2060 nd->last.name = name;
2061 nd->last_type = type;
2063 name += hashlen_len(hash_len);
2067 * If it wasn't NUL, we know it was '/'. Skip that
2068 * slash, and continue until no more slashes.
2072 } while (unlikely(*name == '/'));
2073 if (unlikely(!*name)) {
2075 /* pathname body, done */
2078 name = nd->stack[nd->depth - 1].name;
2079 /* trailing symlink, done */
2082 /* last component of nested symlink */
2083 err = walk_component(nd, WALK_GET | WALK_PUT);
2085 err = walk_component(nd, WALK_GET);
2091 const char *s = get_link(nd);
2100 nd->stack[nd->depth - 1].name = name;
2105 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2106 if (nd->flags & LOOKUP_RCU) {
2107 if (unlazy_walk(nd, NULL, 0))
2115 static const char *path_init(struct nameidata *nd, unsigned flags)
2118 const char *s = nd->name->name;
2120 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2121 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2123 if (flags & LOOKUP_ROOT) {
2124 struct dentry *root = nd->root.dentry;
2125 struct inode *inode = root->d_inode;
2127 if (!d_can_lookup(root))
2128 return ERR_PTR(-ENOTDIR);
2129 retval = inode_permission(inode, MAY_EXEC);
2131 return ERR_PTR(retval);
2133 nd->path = nd->root;
2135 if (flags & LOOKUP_RCU) {
2137 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2138 nd->root_seq = nd->seq;
2139 nd->m_seq = read_seqbegin(&mount_lock);
2141 path_get(&nd->path);
2146 nd->root.mnt = NULL;
2147 nd->path.mnt = NULL;
2148 nd->path.dentry = NULL;
2150 nd->m_seq = read_seqbegin(&mount_lock);
2152 if (flags & LOOKUP_RCU)
2155 if (likely(!nd_jump_root(nd)))
2157 nd->root.mnt = NULL;
2159 return ERR_PTR(-ECHILD);
2160 } else if (nd->dfd == AT_FDCWD) {
2161 if (flags & LOOKUP_RCU) {
2162 struct fs_struct *fs = current->fs;
2168 seq = read_seqcount_begin(&fs->seq);
2170 nd->inode = nd->path.dentry->d_inode;
2171 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2172 } while (read_seqcount_retry(&fs->seq, seq));
2174 get_fs_pwd(current->fs, &nd->path);
2175 nd->inode = nd->path.dentry->d_inode;
2179 /* Caller must check execute permissions on the starting path component */
2180 struct fd f = fdget_raw(nd->dfd);
2181 struct dentry *dentry;
2184 return ERR_PTR(-EBADF);
2186 dentry = f.file->f_path.dentry;
2189 if (!d_can_lookup(dentry)) {
2191 return ERR_PTR(-ENOTDIR);
2195 nd->path = f.file->f_path;
2196 if (flags & LOOKUP_RCU) {
2198 nd->inode = nd->path.dentry->d_inode;
2199 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2201 path_get(&nd->path);
2202 nd->inode = nd->path.dentry->d_inode;
2209 static const char *trailing_symlink(struct nameidata *nd)
2212 int error = may_follow_link(nd);
2213 if (unlikely(error))
2214 return ERR_PTR(error);
2215 nd->flags |= LOOKUP_PARENT;
2216 nd->stack[0].name = NULL;
2221 static inline int lookup_last(struct nameidata *nd)
2223 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2224 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2226 nd->flags &= ~LOOKUP_PARENT;
2227 return walk_component(nd,
2228 nd->flags & LOOKUP_FOLLOW
2230 ? WALK_PUT | WALK_GET
2235 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2236 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2238 const char *s = path_init(nd, flags);
2243 while (!(err = link_path_walk(s, nd))
2244 && ((err = lookup_last(nd)) > 0)) {
2245 s = trailing_symlink(nd);
2252 err = complete_walk(nd);
2254 if (!err && nd->flags & LOOKUP_DIRECTORY)
2255 if (!d_can_lookup(nd->path.dentry))
2259 nd->path.mnt = NULL;
2260 nd->path.dentry = NULL;
2266 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2267 struct path *path, struct path *root)
2270 struct nameidata nd;
2272 return PTR_ERR(name);
2273 if (unlikely(root)) {
2275 flags |= LOOKUP_ROOT;
2277 set_nameidata(&nd, dfd, name);
2278 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2279 if (unlikely(retval == -ECHILD))
2280 retval = path_lookupat(&nd, flags, path);
2281 if (unlikely(retval == -ESTALE))
2282 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2284 if (likely(!retval))
2285 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2286 restore_nameidata();
2291 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2292 static int path_parentat(struct nameidata *nd, unsigned flags,
2293 struct path *parent)
2295 const char *s = path_init(nd, flags);
2299 err = link_path_walk(s, nd);
2301 err = complete_walk(nd);
2304 nd->path.mnt = NULL;
2305 nd->path.dentry = NULL;
2311 static struct filename *filename_parentat(int dfd, struct filename *name,
2312 unsigned int flags, struct path *parent,
2313 struct qstr *last, int *type)
2316 struct nameidata nd;
2320 set_nameidata(&nd, dfd, name);
2321 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2322 if (unlikely(retval == -ECHILD))
2323 retval = path_parentat(&nd, flags, parent);
2324 if (unlikely(retval == -ESTALE))
2325 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2326 if (likely(!retval)) {
2328 *type = nd.last_type;
2329 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2332 name = ERR_PTR(retval);
2334 restore_nameidata();
2338 /* does lookup, returns the object with parent locked */
2339 struct dentry *kern_path_locked(const char *name, struct path *path)
2341 struct filename *filename;
2346 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2348 if (IS_ERR(filename))
2349 return ERR_CAST(filename);
2350 if (unlikely(type != LAST_NORM)) {
2353 return ERR_PTR(-EINVAL);
2355 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2356 d = __lookup_hash(&last, path->dentry, 0);
2358 inode_unlock(path->dentry->d_inode);
2365 int kern_path(const char *name, unsigned int flags, struct path *path)
2367 return filename_lookup(AT_FDCWD, getname_kernel(name),
2370 EXPORT_SYMBOL(kern_path);
2373 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2374 * @dentry: pointer to dentry of the base directory
2375 * @mnt: pointer to vfs mount of the base directory
2376 * @name: pointer to file name
2377 * @flags: lookup flags
2378 * @path: pointer to struct path to fill
2380 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2381 const char *name, unsigned int flags,
2384 struct path root = {.mnt = mnt, .dentry = dentry};
2385 /* the first argument of filename_lookup() is ignored with root */
2386 return filename_lookup(AT_FDCWD, getname_kernel(name),
2387 flags , path, &root);
2389 EXPORT_SYMBOL(vfs_path_lookup);
2392 * lookup_hash - lookup single pathname component on already hashed name
2393 * @name: name and hash to lookup
2394 * @base: base directory to lookup from
2396 * The name must have been verified and hashed (see lookup_one_len()). Using
2397 * this after just full_name_hash() is unsafe.
2399 * This function also doesn't check for search permission on base directory.
2401 * Use lookup_one_len_unlocked() instead, unless you really know what you are
2404 * Do not hold i_mutex; this helper takes i_mutex if necessary.
2406 struct dentry *lookup_hash(const struct qstr *name, struct dentry *base)
2410 ret = lookup_dcache(name, base, 0);
2412 ret = lookup_slow(name, base, 0);
2416 EXPORT_SYMBOL(lookup_hash);
2419 * lookup_one_len - filesystem helper to lookup single pathname component
2420 * @name: pathname component to lookup
2421 * @base: base directory to lookup from
2422 * @len: maximum length @len should be interpreted to
2424 * Note that this routine is purely a helper for filesystem usage and should
2425 * not be called by generic code.
2427 * The caller must hold base->i_mutex.
2429 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2435 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2439 this.hash = full_name_hash(name, len);
2441 return ERR_PTR(-EACCES);
2443 if (unlikely(name[0] == '.')) {
2444 if (len < 2 || (len == 2 && name[1] == '.'))
2445 return ERR_PTR(-EACCES);
2449 c = *(const unsigned char *)name++;
2450 if (c == '/' || c == '\0')
2451 return ERR_PTR(-EACCES);
2454 * See if the low-level filesystem might want
2455 * to use its own hash..
2457 if (base->d_flags & DCACHE_OP_HASH) {
2458 int err = base->d_op->d_hash(base, &this);
2460 return ERR_PTR(err);
2463 err = inode_permission(base->d_inode, MAY_EXEC);
2465 return ERR_PTR(err);
2467 return __lookup_hash(&this, base, 0);
2469 EXPORT_SYMBOL(lookup_one_len);
2472 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2473 * @name: pathname component to lookup
2474 * @base: base directory to lookup from
2475 * @len: maximum length @len should be interpreted to
2477 * Note that this routine is purely a helper for filesystem usage and should
2478 * not be called by generic code.
2480 * Unlike lookup_one_len, it should be called without the parent
2481 * i_mutex held, and will take the i_mutex itself if necessary.
2483 struct dentry *lookup_one_len_unlocked(const char *name,
2484 struct dentry *base, int len)
2492 this.hash = full_name_hash(name, len);
2494 return ERR_PTR(-EACCES);
2496 if (unlikely(name[0] == '.')) {
2497 if (len < 2 || (len == 2 && name[1] == '.'))
2498 return ERR_PTR(-EACCES);
2502 c = *(const unsigned char *)name++;
2503 if (c == '/' || c == '\0')
2504 return ERR_PTR(-EACCES);
2507 * See if the low-level filesystem might want
2508 * to use its own hash..
2510 if (base->d_flags & DCACHE_OP_HASH) {
2511 int err = base->d_op->d_hash(base, &this);
2513 return ERR_PTR(err);
2516 err = inode_permission(base->d_inode, MAY_EXEC);
2518 return ERR_PTR(err);
2520 return lookup_hash(&this, base);
2522 EXPORT_SYMBOL(lookup_one_len_unlocked);
2524 #ifdef CONFIG_UNIX98_PTYS
2525 int path_pts(struct path *path)
2527 /* Find something mounted on "pts" in the same directory as
2530 struct dentry *child, *parent;
2534 ret = path_parent_directory(path);
2538 parent = path->dentry;
2541 child = d_hash_and_lookup(parent, &this);
2545 path->dentry = child;
2552 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2553 struct path *path, int *empty)
2555 return filename_lookup(dfd, getname_flags(name, flags, empty),
2558 EXPORT_SYMBOL(user_path_at_empty);
2561 * NB: most callers don't do anything directly with the reference to the
2562 * to struct filename, but the nd->last pointer points into the name string
2563 * allocated by getname. So we must hold the reference to it until all
2564 * path-walking is complete.
2566 static inline struct filename *
2567 user_path_parent(int dfd, const char __user *path,
2568 struct path *parent,
2573 /* only LOOKUP_REVAL is allowed in extra flags */
2574 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2575 parent, last, type);
2579 * mountpoint_last - look up last component for umount
2580 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2581 * @path: pointer to container for result
2583 * This is a special lookup_last function just for umount. In this case, we
2584 * need to resolve the path without doing any revalidation.
2586 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2587 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2588 * in almost all cases, this lookup will be served out of the dcache. The only
2589 * cases where it won't are if nd->last refers to a symlink or the path is
2590 * bogus and it doesn't exist.
2593 * -error: if there was an error during lookup. This includes -ENOENT if the
2594 * lookup found a negative dentry. The nd->path reference will also be
2597 * 0: if we successfully resolved nd->path and found it to not to be a
2598 * symlink that needs to be followed. "path" will also be populated.
2599 * The nd->path reference will also be put.
2601 * 1: if we successfully resolved nd->last and found it to be a symlink
2602 * that needs to be followed. "path" will be populated with the path
2603 * to the link, and nd->path will *not* be put.
2606 mountpoint_last(struct nameidata *nd, struct path *path)
2609 struct dentry *dentry;
2610 struct dentry *dir = nd->path.dentry;
2612 /* If we're in rcuwalk, drop out of it to handle last component */
2613 if (nd->flags & LOOKUP_RCU) {
2614 if (unlazy_walk(nd, NULL, 0))
2618 nd->flags &= ~LOOKUP_PARENT;
2620 if (unlikely(nd->last_type != LAST_NORM)) {
2621 error = handle_dots(nd, nd->last_type);
2624 dentry = dget(nd->path.dentry);
2626 dentry = d_lookup(dir, &nd->last);
2629 * No cached dentry. Mounted dentries are pinned in the
2630 * cache, so that means that this dentry is probably
2631 * a symlink or the path doesn't actually point
2632 * to a mounted dentry.
2634 dentry = lookup_slow(&nd->last, dir,
2635 nd->flags | LOOKUP_NO_REVAL);
2637 return PTR_ERR(dentry);
2640 if (d_is_negative(dentry)) {
2646 path->dentry = dentry;
2647 path->mnt = nd->path.mnt;
2648 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2649 d_backing_inode(dentry), 0);
2650 if (unlikely(error))
2658 * path_mountpoint - look up a path to be umounted
2659 * @nd: lookup context
2660 * @flags: lookup flags
2661 * @path: pointer to container for result
2663 * Look up the given name, but don't attempt to revalidate the last component.
2664 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2667 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2669 const char *s = path_init(nd, flags);
2673 while (!(err = link_path_walk(s, nd)) &&
2674 (err = mountpoint_last(nd, path)) > 0) {
2675 s = trailing_symlink(nd);
2686 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2689 struct nameidata nd;
2692 return PTR_ERR(name);
2693 set_nameidata(&nd, dfd, name);
2694 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2695 if (unlikely(error == -ECHILD))
2696 error = path_mountpoint(&nd, flags, path);
2697 if (unlikely(error == -ESTALE))
2698 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2700 audit_inode(name, path->dentry, 0);
2701 restore_nameidata();
2707 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2708 * @dfd: directory file descriptor
2709 * @name: pathname from userland
2710 * @flags: lookup flags
2711 * @path: pointer to container to hold result
2713 * A umount is a special case for path walking. We're not actually interested
2714 * in the inode in this situation, and ESTALE errors can be a problem. We
2715 * simply want track down the dentry and vfsmount attached at the mountpoint
2716 * and avoid revalidating the last component.
2718 * Returns 0 and populates "path" on success.
2721 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2724 return filename_mountpoint(dfd, getname(name), path, flags);
2728 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2731 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2733 EXPORT_SYMBOL(kern_path_mountpoint);
2735 int __check_sticky(struct inode *dir, struct inode *inode)
2737 kuid_t fsuid = current_fsuid();
2739 if (uid_eq(inode->i_uid, fsuid))
2741 if (uid_eq(dir->i_uid, fsuid))
2743 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2745 EXPORT_SYMBOL(__check_sticky);
2748 * Check whether we can remove a link victim from directory dir, check
2749 * whether the type of victim is right.
2750 * 1. We can't do it if dir is read-only (done in permission())
2751 * 2. We should have write and exec permissions on dir
2752 * 3. We can't remove anything from append-only dir
2753 * 4. We can't do anything with immutable dir (done in permission())
2754 * 5. If the sticky bit on dir is set we should either
2755 * a. be owner of dir, or
2756 * b. be owner of victim, or
2757 * c. have CAP_FOWNER capability
2758 * 6. If the victim is append-only or immutable we can't do antyhing with
2759 * links pointing to it.
2760 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2761 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2762 * 9. We can't remove a root or mountpoint.
2763 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2764 * nfs_async_unlink().
2766 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2768 struct inode *inode = d_backing_inode(victim);
2771 if (d_is_negative(victim))
2775 BUG_ON(victim->d_parent->d_inode != dir);
2776 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2778 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2784 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2785 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2788 if (!d_is_dir(victim))
2790 if (IS_ROOT(victim))
2792 } else if (d_is_dir(victim))
2794 if (IS_DEADDIR(dir))
2796 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2801 /* Check whether we can create an object with dentry child in directory
2803 * 1. We can't do it if child already exists (open has special treatment for
2804 * this case, but since we are inlined it's OK)
2805 * 2. We can't do it if dir is read-only (done in permission())
2806 * 3. We should have write and exec permissions on dir
2807 * 4. We can't do it if dir is immutable (done in permission())
2809 static inline int may_create(struct inode *dir, struct dentry *child)
2811 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2814 if (IS_DEADDIR(dir))
2816 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2820 * p1 and p2 should be directories on the same fs.
2822 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2827 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2831 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2833 p = d_ancestor(p2, p1);
2835 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2836 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2840 p = d_ancestor(p1, p2);
2842 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2843 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2847 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2848 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2851 EXPORT_SYMBOL(lock_rename);
2853 void unlock_rename(struct dentry *p1, struct dentry *p2)
2855 inode_unlock(p1->d_inode);
2857 inode_unlock(p2->d_inode);
2858 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2861 EXPORT_SYMBOL(unlock_rename);
2863 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2866 int error = may_create(dir, dentry);
2870 if (!dir->i_op->create)
2871 return -EACCES; /* shouldn't it be ENOSYS? */
2874 error = security_inode_create(dir, dentry, mode);
2877 error = dir->i_op->create(dir, dentry, mode, want_excl);
2879 fsnotify_create(dir, dentry);
2882 EXPORT_SYMBOL(vfs_create);
2884 static int may_open(struct path *path, int acc_mode, int flag)
2886 struct dentry *dentry = path->dentry;
2887 struct inode *inode = dentry->d_inode;
2893 switch (inode->i_mode & S_IFMT) {
2897 if (acc_mode & MAY_WRITE)
2902 if (path->mnt->mnt_flags & MNT_NODEV)
2911 error = inode_permission(inode, MAY_OPEN | acc_mode);
2916 * An append-only file must be opened in append mode for writing.
2918 if (IS_APPEND(inode)) {
2919 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2925 /* O_NOATIME can only be set by the owner or superuser */
2926 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2932 static int handle_truncate(struct file *filp)
2934 struct path *path = &filp->f_path;
2935 struct inode *inode = path->dentry->d_inode;
2936 int error = get_write_access(inode);
2940 * Refuse to truncate files with mandatory locks held on them.
2942 error = locks_verify_locked(filp);
2944 error = security_path_truncate(path);
2946 error = do_truncate(path->dentry, 0,
2947 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2950 put_write_access(inode);
2954 static inline int open_to_namei_flags(int flag)
2956 if ((flag & O_ACCMODE) == 3)
2961 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2963 int error = security_path_mknod(dir, dentry, mode, 0);
2967 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2971 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2975 * Attempt to atomically look up, create and open a file from a negative
2978 * Returns 0 if successful. The file will have been created and attached to
2979 * @file by the filesystem calling finish_open().
2981 * Returns 1 if the file was looked up only or didn't need creating. The
2982 * caller will need to perform the open themselves. @path will have been
2983 * updated to point to the new dentry. This may be negative.
2985 * Returns an error code otherwise.
2987 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2988 struct path *path, struct file *file,
2989 const struct open_flags *op,
2990 int open_flag, umode_t mode,
2993 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2994 struct inode *dir = nd->path.dentry->d_inode;
2997 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
2998 open_flag &= ~O_TRUNC;
3000 if (nd->flags & LOOKUP_DIRECTORY)
3001 open_flag |= O_DIRECTORY;
3003 file->f_path.dentry = DENTRY_NOT_SET;
3004 file->f_path.mnt = nd->path.mnt;
3005 error = dir->i_op->atomic_open(dir, dentry, file,
3006 open_to_namei_flags(open_flag),
3008 d_lookup_done(dentry);
3011 * We didn't have the inode before the open, so check open
3014 int acc_mode = op->acc_mode;
3015 if (*opened & FILE_CREATED) {
3016 WARN_ON(!(open_flag & O_CREAT));
3017 fsnotify_create(dir, dentry);
3020 error = may_open(&file->f_path, acc_mode, open_flag);
3021 if (WARN_ON(error > 0))
3023 } else if (error > 0) {
3024 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3027 if (file->f_path.dentry) {
3029 dentry = file->f_path.dentry;
3031 if (*opened & FILE_CREATED)
3032 fsnotify_create(dir, dentry);
3033 if (unlikely(d_is_negative(dentry))) {
3036 path->dentry = dentry;
3037 path->mnt = nd->path.mnt;
3047 * Look up and maybe create and open the last component.
3049 * Must be called with i_mutex held on parent.
3051 * Returns 0 if the file was successfully atomically created (if necessary) and
3052 * opened. In this case the file will be returned attached to @file.
3054 * Returns 1 if the file was not completely opened at this time, though lookups
3055 * and creations will have been performed and the dentry returned in @path will
3056 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3057 * specified then a negative dentry may be returned.
3059 * An error code is returned otherwise.
3061 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3062 * cleared otherwise prior to returning.
3064 static int lookup_open(struct nameidata *nd, struct path *path,
3066 const struct open_flags *op,
3067 bool got_write, int *opened)
3069 struct dentry *dir = nd->path.dentry;
3070 struct inode *dir_inode = dir->d_inode;
3071 int open_flag = op->open_flag;
3072 struct dentry *dentry;
3073 int error, create_error = 0;
3074 umode_t mode = op->mode;
3075 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3077 if (unlikely(IS_DEADDIR(dir_inode)))
3080 *opened &= ~FILE_CREATED;
3081 dentry = d_lookup(dir, &nd->last);
3084 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3086 return PTR_ERR(dentry);
3088 if (d_in_lookup(dentry))
3091 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3094 error = d_revalidate(dentry, nd->flags);
3095 if (likely(error > 0))
3099 d_invalidate(dentry);
3103 if (dentry->d_inode) {
3104 /* Cached positive dentry: will open in f_op->open */
3109 * Checking write permission is tricky, bacuse we don't know if we are
3110 * going to actually need it: O_CREAT opens should work as long as the
3111 * file exists. But checking existence breaks atomicity. The trick is
3112 * to check access and if not granted clear O_CREAT from the flags.
3114 * Another problem is returing the "right" error value (e.g. for an
3115 * O_EXCL open we want to return EEXIST not EROFS).
3117 if (open_flag & O_CREAT) {
3118 if (!IS_POSIXACL(dir->d_inode))
3119 mode &= ~current_umask();
3120 if (unlikely(!got_write)) {
3121 create_error = -EROFS;
3122 open_flag &= ~O_CREAT;
3123 if (open_flag & (O_EXCL | O_TRUNC))
3125 /* No side effects, safe to clear O_CREAT */
3127 create_error = may_o_create(&nd->path, dentry, mode);
3129 open_flag &= ~O_CREAT;
3130 if (open_flag & O_EXCL)
3134 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3135 unlikely(!got_write)) {
3137 * No O_CREATE -> atomicity not a requirement -> fall
3138 * back to lookup + open
3143 if (dir_inode->i_op->atomic_open) {
3144 error = atomic_open(nd, dentry, path, file, op, open_flag,
3146 if (unlikely(error == -ENOENT) && create_error)
3147 error = create_error;
3152 if (d_in_lookup(dentry)) {
3153 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3155 d_lookup_done(dentry);
3156 if (unlikely(res)) {
3158 error = PTR_ERR(res);
3166 /* Negative dentry, just create the file */
3167 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3168 *opened |= FILE_CREATED;
3169 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3170 if (!dir_inode->i_op->create) {
3174 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3175 open_flag & O_EXCL);
3178 fsnotify_create(dir_inode, dentry);
3180 if (unlikely(create_error) && !dentry->d_inode) {
3181 error = create_error;
3185 path->dentry = dentry;
3186 path->mnt = nd->path.mnt;
3195 * Handle the last step of open()
3197 static int do_last(struct nameidata *nd,
3198 struct file *file, const struct open_flags *op,
3201 struct dentry *dir = nd->path.dentry;
3202 int open_flag = op->open_flag;
3203 bool will_truncate = (open_flag & O_TRUNC) != 0;
3204 bool got_write = false;
3205 int acc_mode = op->acc_mode;
3207 struct inode *inode;
3211 nd->flags &= ~LOOKUP_PARENT;
3212 nd->flags |= op->intent;
3214 if (nd->last_type != LAST_NORM) {
3215 error = handle_dots(nd, nd->last_type);
3216 if (unlikely(error))
3221 if (!(open_flag & O_CREAT)) {
3222 if (nd->last.name[nd->last.len])
3223 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3224 /* we _can_ be in RCU mode here */
3225 error = lookup_fast(nd, &path, &inode, &seq);
3226 if (likely(error > 0))
3232 BUG_ON(nd->inode != dir->d_inode);
3233 BUG_ON(nd->flags & LOOKUP_RCU);
3235 /* create side of things */
3237 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3238 * has been cleared when we got to the last component we are
3241 error = complete_walk(nd);
3245 audit_inode(nd->name, dir, LOOKUP_PARENT);
3246 /* trailing slashes? */
3247 if (unlikely(nd->last.name[nd->last.len]))
3251 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3252 error = mnt_want_write(nd->path.mnt);
3256 * do _not_ fail yet - we might not need that or fail with
3257 * a different error; let lookup_open() decide; we'll be
3258 * dropping this one anyway.
3261 if (open_flag & O_CREAT)
3262 inode_lock(dir->d_inode);
3264 inode_lock_shared(dir->d_inode);
3265 error = lookup_open(nd, &path, file, op, got_write, opened);
3266 if (open_flag & O_CREAT)
3267 inode_unlock(dir->d_inode);
3269 inode_unlock_shared(dir->d_inode);
3275 if ((*opened & FILE_CREATED) ||
3276 !S_ISREG(file_inode(file)->i_mode))
3277 will_truncate = false;
3279 audit_inode(nd->name, file->f_path.dentry, 0);
3283 if (*opened & FILE_CREATED) {
3284 /* Don't check for write permission, don't truncate */
3285 open_flag &= ~O_TRUNC;
3286 will_truncate = false;
3288 path_to_nameidata(&path, nd);
3289 goto finish_open_created;
3293 * If atomic_open() acquired write access it is dropped now due to
3294 * possible mount and symlink following (this might be optimized away if
3298 mnt_drop_write(nd->path.mnt);
3302 error = follow_managed(&path, nd);
3303 if (unlikely(error < 0))
3306 if (unlikely(d_is_negative(path.dentry))) {
3307 path_to_nameidata(&path, nd);
3312 * create/update audit record if it already exists.
3314 audit_inode(nd->name, path.dentry, 0);
3316 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3317 path_to_nameidata(&path, nd);
3321 seq = 0; /* out of RCU mode, so the value doesn't matter */
3322 inode = d_backing_inode(path.dentry);
3326 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3328 if (unlikely(error))
3331 path_to_nameidata(&path, nd);
3334 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3336 error = complete_walk(nd);
3339 audit_inode(nd->name, nd->path.dentry, 0);
3341 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3344 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3346 if (!d_is_reg(nd->path.dentry))
3347 will_truncate = false;
3349 if (will_truncate) {
3350 error = mnt_want_write(nd->path.mnt);
3355 finish_open_created:
3356 error = may_open(&nd->path, acc_mode, open_flag);
3359 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3360 error = vfs_open(&nd->path, file, current_cred());
3363 *opened |= FILE_OPENED;
3365 error = open_check_o_direct(file);
3367 error = ima_file_check(file, op->acc_mode, *opened);
3368 if (!error && will_truncate)
3369 error = handle_truncate(file);
3371 if (unlikely(error) && (*opened & FILE_OPENED))
3373 if (unlikely(error > 0)) {
3378 mnt_drop_write(nd->path.mnt);
3382 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3383 const struct open_flags *op,
3384 struct file *file, int *opened)
3386 static const struct qstr name = QSTR_INIT("/", 1);
3387 struct dentry *child;
3390 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3391 if (unlikely(error))
3393 error = mnt_want_write(path.mnt);
3394 if (unlikely(error))
3396 dir = path.dentry->d_inode;
3397 /* we want directory to be writable */
3398 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3401 if (!dir->i_op->tmpfile) {
3402 error = -EOPNOTSUPP;
3405 child = d_alloc(path.dentry, &name);
3406 if (unlikely(!child)) {
3411 path.dentry = child;
3412 error = dir->i_op->tmpfile(dir, child, op->mode);
3415 audit_inode(nd->name, child, 0);
3416 /* Don't check for other permissions, the inode was just created */
3417 error = may_open(&path, 0, op->open_flag);
3420 file->f_path.mnt = path.mnt;
3421 error = finish_open(file, child, NULL, opened);
3424 error = open_check_o_direct(file);
3427 } else if (!(op->open_flag & O_EXCL)) {
3428 struct inode *inode = file_inode(file);
3429 spin_lock(&inode->i_lock);
3430 inode->i_state |= I_LINKABLE;
3431 spin_unlock(&inode->i_lock);
3434 mnt_drop_write(path.mnt);
3440 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3443 int error = path_lookupat(nd, flags, &path);
3445 audit_inode(nd->name, path.dentry, 0);
3446 error = vfs_open(&path, file, current_cred());
3452 static struct file *path_openat(struct nameidata *nd,
3453 const struct open_flags *op, unsigned flags)
3460 file = get_empty_filp();
3464 file->f_flags = op->open_flag;
3466 if (unlikely(file->f_flags & __O_TMPFILE)) {
3467 error = do_tmpfile(nd, flags, op, file, &opened);
3471 if (unlikely(file->f_flags & O_PATH)) {
3472 error = do_o_path(nd, flags, file);
3474 opened |= FILE_OPENED;
3478 s = path_init(nd, flags);
3483 while (!(error = link_path_walk(s, nd)) &&
3484 (error = do_last(nd, file, op, &opened)) > 0) {
3485 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3486 s = trailing_symlink(nd);
3494 if (!(opened & FILE_OPENED)) {
3498 if (unlikely(error)) {
3499 if (error == -EOPENSTALE) {
3500 if (flags & LOOKUP_RCU)
3505 file = ERR_PTR(error);
3510 struct file *do_filp_open(int dfd, struct filename *pathname,
3511 const struct open_flags *op)
3513 struct nameidata nd;
3514 int flags = op->lookup_flags;
3517 set_nameidata(&nd, dfd, pathname);
3518 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3519 if (unlikely(filp == ERR_PTR(-ECHILD)))
3520 filp = path_openat(&nd, op, flags);
3521 if (unlikely(filp == ERR_PTR(-ESTALE)))
3522 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3523 restore_nameidata();
3527 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3528 const char *name, const struct open_flags *op)
3530 struct nameidata nd;
3532 struct filename *filename;
3533 int flags = op->lookup_flags | LOOKUP_ROOT;
3536 nd.root.dentry = dentry;
3538 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3539 return ERR_PTR(-ELOOP);
3541 filename = getname_kernel(name);
3542 if (IS_ERR(filename))
3543 return ERR_CAST(filename);
3545 set_nameidata(&nd, -1, filename);
3546 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3547 if (unlikely(file == ERR_PTR(-ECHILD)))
3548 file = path_openat(&nd, op, flags);
3549 if (unlikely(file == ERR_PTR(-ESTALE)))
3550 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3551 restore_nameidata();
3556 static struct dentry *filename_create(int dfd, struct filename *name,
3557 struct path *path, unsigned int lookup_flags)
3559 struct dentry *dentry = ERR_PTR(-EEXIST);
3564 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3567 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3568 * other flags passed in are ignored!
3570 lookup_flags &= LOOKUP_REVAL;
3572 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3574 return ERR_CAST(name);
3577 * Yucky last component or no last component at all?
3578 * (foo/., foo/.., /////)
3580 if (unlikely(type != LAST_NORM))
3583 /* don't fail immediately if it's r/o, at least try to report other errors */
3584 err2 = mnt_want_write(path->mnt);
3586 * Do the final lookup.
3588 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3589 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3590 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3595 if (d_is_positive(dentry))
3599 * Special case - lookup gave negative, but... we had foo/bar/
3600 * From the vfs_mknod() POV we just have a negative dentry -
3601 * all is fine. Let's be bastards - you had / on the end, you've
3602 * been asking for (non-existent) directory. -ENOENT for you.
3604 if (unlikely(!is_dir && last.name[last.len])) {
3608 if (unlikely(err2)) {
3616 dentry = ERR_PTR(error);
3618 inode_unlock(path->dentry->d_inode);
3620 mnt_drop_write(path->mnt);
3627 struct dentry *kern_path_create(int dfd, const char *pathname,
3628 struct path *path, unsigned int lookup_flags)
3630 return filename_create(dfd, getname_kernel(pathname),
3631 path, lookup_flags);
3633 EXPORT_SYMBOL(kern_path_create);
3635 void done_path_create(struct path *path, struct dentry *dentry)
3638 inode_unlock(path->dentry->d_inode);
3639 mnt_drop_write(path->mnt);
3642 EXPORT_SYMBOL(done_path_create);
3644 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3645 struct path *path, unsigned int lookup_flags)
3647 return filename_create(dfd, getname(pathname), path, lookup_flags);
3649 EXPORT_SYMBOL(user_path_create);
3651 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3653 int error = may_create(dir, dentry);
3658 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3661 if (!dir->i_op->mknod)
3664 error = devcgroup_inode_mknod(mode, dev);
3668 error = security_inode_mknod(dir, dentry, mode, dev);
3672 error = dir->i_op->mknod(dir, dentry, mode, dev);
3674 fsnotify_create(dir, dentry);
3677 EXPORT_SYMBOL(vfs_mknod);
3679 static int may_mknod(umode_t mode)
3681 switch (mode & S_IFMT) {
3687 case 0: /* zero mode translates to S_IFREG */
3696 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3699 struct dentry *dentry;
3702 unsigned int lookup_flags = 0;
3704 error = may_mknod(mode);
3708 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3710 return PTR_ERR(dentry);
3712 if (!IS_POSIXACL(path.dentry->d_inode))
3713 mode &= ~current_umask();
3714 error = security_path_mknod(&path, dentry, mode, dev);
3717 switch (mode & S_IFMT) {
3718 case 0: case S_IFREG:
3719 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3721 ima_post_path_mknod(dentry);
3723 case S_IFCHR: case S_IFBLK:
3724 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3725 new_decode_dev(dev));
3727 case S_IFIFO: case S_IFSOCK:
3728 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3732 done_path_create(&path, dentry);
3733 if (retry_estale(error, lookup_flags)) {
3734 lookup_flags |= LOOKUP_REVAL;
3740 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3742 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3745 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3747 int error = may_create(dir, dentry);
3748 unsigned max_links = dir->i_sb->s_max_links;
3753 if (!dir->i_op->mkdir)
3756 mode &= (S_IRWXUGO|S_ISVTX);
3757 error = security_inode_mkdir(dir, dentry, mode);
3761 if (max_links && dir->i_nlink >= max_links)
3764 error = dir->i_op->mkdir(dir, dentry, mode);
3766 fsnotify_mkdir(dir, dentry);
3769 EXPORT_SYMBOL(vfs_mkdir);
3771 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3773 struct dentry *dentry;
3776 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3779 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3781 return PTR_ERR(dentry);
3783 if (!IS_POSIXACL(path.dentry->d_inode))
3784 mode &= ~current_umask();
3785 error = security_path_mkdir(&path, dentry, mode);
3787 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3788 done_path_create(&path, dentry);
3789 if (retry_estale(error, lookup_flags)) {
3790 lookup_flags |= LOOKUP_REVAL;
3796 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3798 return sys_mkdirat(AT_FDCWD, pathname, mode);
3801 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3803 int error = may_delete(dir, dentry, 1);
3808 if (!dir->i_op->rmdir)
3812 inode_lock(dentry->d_inode);
3815 if (is_local_mountpoint(dentry))
3818 error = security_inode_rmdir(dir, dentry);
3822 shrink_dcache_parent(dentry);
3823 error = dir->i_op->rmdir(dir, dentry);
3827 dentry->d_inode->i_flags |= S_DEAD;
3829 detach_mounts(dentry);
3832 inode_unlock(dentry->d_inode);
3838 EXPORT_SYMBOL(vfs_rmdir);
3840 static long do_rmdir(int dfd, const char __user *pathname)
3843 struct filename *name;
3844 struct dentry *dentry;
3848 unsigned int lookup_flags = 0;
3850 name = user_path_parent(dfd, pathname,
3851 &path, &last, &type, lookup_flags);
3853 return PTR_ERR(name);
3867 error = mnt_want_write(path.mnt);
3871 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3872 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3873 error = PTR_ERR(dentry);
3876 if (!dentry->d_inode) {
3880 error = security_path_rmdir(&path, dentry);
3883 error = vfs_rmdir(path.dentry->d_inode, dentry);
3887 inode_unlock(path.dentry->d_inode);
3888 mnt_drop_write(path.mnt);
3892 if (retry_estale(error, lookup_flags)) {
3893 lookup_flags |= LOOKUP_REVAL;
3899 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3901 return do_rmdir(AT_FDCWD, pathname);
3905 * vfs_unlink - unlink a filesystem object
3906 * @dir: parent directory
3908 * @delegated_inode: returns victim inode, if the inode is delegated.
3910 * The caller must hold dir->i_mutex.
3912 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3913 * return a reference to the inode in delegated_inode. The caller
3914 * should then break the delegation on that inode and retry. Because
3915 * breaking a delegation may take a long time, the caller should drop
3916 * dir->i_mutex before doing so.
3918 * Alternatively, a caller may pass NULL for delegated_inode. This may
3919 * be appropriate for callers that expect the underlying filesystem not
3920 * to be NFS exported.
3922 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3924 struct inode *target = dentry->d_inode;
3925 int error = may_delete(dir, dentry, 0);
3930 if (!dir->i_op->unlink)
3934 if (is_local_mountpoint(dentry))
3937 error = security_inode_unlink(dir, dentry);
3939 error = try_break_deleg(target, delegated_inode);
3942 error = dir->i_op->unlink(dir, dentry);
3945 detach_mounts(dentry);
3950 inode_unlock(target);
3952 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3953 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3954 fsnotify_link_count(target);
3960 EXPORT_SYMBOL(vfs_unlink);
3963 * Make sure that the actual truncation of the file will occur outside its
3964 * directory's i_mutex. Truncate can take a long time if there is a lot of
3965 * writeout happening, and we don't want to prevent access to the directory
3966 * while waiting on the I/O.
3968 static long do_unlinkat(int dfd, const char __user *pathname)
3971 struct filename *name;
3972 struct dentry *dentry;
3976 struct inode *inode = NULL;
3977 struct inode *delegated_inode = NULL;
3978 unsigned int lookup_flags = 0;
3980 name = user_path_parent(dfd, pathname,
3981 &path, &last, &type, lookup_flags);
3983 return PTR_ERR(name);
3986 if (type != LAST_NORM)
3989 error = mnt_want_write(path.mnt);
3993 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3994 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3995 error = PTR_ERR(dentry);
3996 if (!IS_ERR(dentry)) {
3997 /* Why not before? Because we want correct error value */
3998 if (last.name[last.len])
4000 inode = dentry->d_inode;
4001 if (d_is_negative(dentry))
4004 error = security_path_unlink(&path, dentry);
4007 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4011 inode_unlock(path.dentry->d_inode);
4013 iput(inode); /* truncate the inode here */
4015 if (delegated_inode) {
4016 error = break_deleg_wait(&delegated_inode);
4020 mnt_drop_write(path.mnt);
4024 if (retry_estale(error, lookup_flags)) {
4025 lookup_flags |= LOOKUP_REVAL;
4032 if (d_is_negative(dentry))
4034 else if (d_is_dir(dentry))
4041 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4043 if ((flag & ~AT_REMOVEDIR) != 0)
4046 if (flag & AT_REMOVEDIR)
4047 return do_rmdir(dfd, pathname);
4049 return do_unlinkat(dfd, pathname);
4052 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4054 return do_unlinkat(AT_FDCWD, pathname);
4057 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4059 int error = may_create(dir, dentry);
4064 if (!dir->i_op->symlink)
4067 error = security_inode_symlink(dir, dentry, oldname);
4071 error = dir->i_op->symlink(dir, dentry, oldname);
4073 fsnotify_create(dir, dentry);
4076 EXPORT_SYMBOL(vfs_symlink);
4078 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4079 int, newdfd, const char __user *, newname)
4082 struct filename *from;
4083 struct dentry *dentry;
4085 unsigned int lookup_flags = 0;
4087 from = getname(oldname);
4089 return PTR_ERR(from);
4091 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4092 error = PTR_ERR(dentry);
4096 error = security_path_symlink(&path, dentry, from->name);
4098 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4099 done_path_create(&path, dentry);
4100 if (retry_estale(error, lookup_flags)) {
4101 lookup_flags |= LOOKUP_REVAL;
4109 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4111 return sys_symlinkat(oldname, AT_FDCWD, newname);
4115 * vfs_link - create a new link
4116 * @old_dentry: object to be linked
4118 * @new_dentry: where to create the new link
4119 * @delegated_inode: returns inode needing a delegation break
4121 * The caller must hold dir->i_mutex
4123 * If vfs_link discovers a delegation on the to-be-linked file in need
4124 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4125 * inode in delegated_inode. The caller should then break the delegation
4126 * and retry. Because breaking a delegation may take a long time, the
4127 * caller should drop the i_mutex before doing so.
4129 * Alternatively, a caller may pass NULL for delegated_inode. This may
4130 * be appropriate for callers that expect the underlying filesystem not
4131 * to be NFS exported.
4133 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4135 struct inode *inode = old_dentry->d_inode;
4136 unsigned max_links = dir->i_sb->s_max_links;
4142 error = may_create(dir, new_dentry);
4146 if (dir->i_sb != inode->i_sb)
4150 * A link to an append-only or immutable file cannot be created.
4152 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4154 if (!dir->i_op->link)
4156 if (S_ISDIR(inode->i_mode))
4159 error = security_inode_link(old_dentry, dir, new_dentry);
4164 /* Make sure we don't allow creating hardlink to an unlinked file */
4165 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4167 else if (max_links && inode->i_nlink >= max_links)
4170 error = try_break_deleg(inode, delegated_inode);
4172 error = dir->i_op->link(old_dentry, dir, new_dentry);
4175 if (!error && (inode->i_state & I_LINKABLE)) {
4176 spin_lock(&inode->i_lock);
4177 inode->i_state &= ~I_LINKABLE;
4178 spin_unlock(&inode->i_lock);
4180 inode_unlock(inode);
4182 fsnotify_link(dir, inode, new_dentry);
4185 EXPORT_SYMBOL(vfs_link);
4188 * Hardlinks are often used in delicate situations. We avoid
4189 * security-related surprises by not following symlinks on the
4192 * We don't follow them on the oldname either to be compatible
4193 * with linux 2.0, and to avoid hard-linking to directories
4194 * and other special files. --ADM
4196 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4197 int, newdfd, const char __user *, newname, int, flags)
4199 struct dentry *new_dentry;
4200 struct path old_path, new_path;
4201 struct inode *delegated_inode = NULL;
4205 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4208 * To use null names we require CAP_DAC_READ_SEARCH
4209 * This ensures that not everyone will be able to create
4210 * handlink using the passed filedescriptor.
4212 if (flags & AT_EMPTY_PATH) {
4213 if (!capable(CAP_DAC_READ_SEARCH))
4218 if (flags & AT_SYMLINK_FOLLOW)
4219 how |= LOOKUP_FOLLOW;
4221 error = user_path_at(olddfd, oldname, how, &old_path);
4225 new_dentry = user_path_create(newdfd, newname, &new_path,
4226 (how & LOOKUP_REVAL));
4227 error = PTR_ERR(new_dentry);
4228 if (IS_ERR(new_dentry))
4232 if (old_path.mnt != new_path.mnt)
4234 error = may_linkat(&old_path);
4235 if (unlikely(error))
4237 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4240 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4242 done_path_create(&new_path, new_dentry);
4243 if (delegated_inode) {
4244 error = break_deleg_wait(&delegated_inode);
4246 path_put(&old_path);
4250 if (retry_estale(error, how)) {
4251 path_put(&old_path);
4252 how |= LOOKUP_REVAL;
4256 path_put(&old_path);
4261 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4263 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4267 * vfs_rename - rename a filesystem object
4268 * @old_dir: parent of source
4269 * @old_dentry: source
4270 * @new_dir: parent of destination
4271 * @new_dentry: destination
4272 * @delegated_inode: returns an inode needing a delegation break
4273 * @flags: rename flags
4275 * The caller must hold multiple mutexes--see lock_rename()).
4277 * If vfs_rename discovers a delegation in need of breaking at either
4278 * the source or destination, it will return -EWOULDBLOCK and return a
4279 * reference to the inode in delegated_inode. The caller should then
4280 * break the delegation and retry. Because breaking a delegation may
4281 * take a long time, the caller should drop all locks before doing
4284 * Alternatively, a caller may pass NULL for delegated_inode. This may
4285 * be appropriate for callers that expect the underlying filesystem not
4286 * to be NFS exported.
4288 * The worst of all namespace operations - renaming directory. "Perverted"
4289 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4291 * a) we can get into loop creation.
4292 * b) race potential - two innocent renames can create a loop together.
4293 * That's where 4.4 screws up. Current fix: serialization on
4294 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4296 * c) we have to lock _four_ objects - parents and victim (if it exists),
4297 * and source (if it is not a directory).
4298 * And that - after we got ->i_mutex on parents (until then we don't know
4299 * whether the target exists). Solution: try to be smart with locking
4300 * order for inodes. We rely on the fact that tree topology may change
4301 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4302 * move will be locked. Thus we can rank directories by the tree
4303 * (ancestors first) and rank all non-directories after them.
4304 * That works since everybody except rename does "lock parent, lookup,
4305 * lock child" and rename is under ->s_vfs_rename_mutex.
4306 * HOWEVER, it relies on the assumption that any object with ->lookup()
4307 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4308 * we'd better make sure that there's no link(2) for them.
4309 * d) conversion from fhandle to dentry may come in the wrong moment - when
4310 * we are removing the target. Solution: we will have to grab ->i_mutex
4311 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4312 * ->i_mutex on parents, which works but leads to some truly excessive
4315 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4316 struct inode *new_dir, struct dentry *new_dentry,
4317 struct inode **delegated_inode, unsigned int flags)
4320 bool is_dir = d_is_dir(old_dentry);
4321 const unsigned char *old_name;
4322 struct inode *source = old_dentry->d_inode;
4323 struct inode *target = new_dentry->d_inode;
4324 bool new_is_dir = false;
4325 unsigned max_links = new_dir->i_sb->s_max_links;
4328 * Check source == target.
4329 * On overlayfs need to look at underlying inodes.
4331 if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4334 error = may_delete(old_dir, old_dentry, is_dir);
4339 error = may_create(new_dir, new_dentry);
4341 new_is_dir = d_is_dir(new_dentry);
4343 if (!(flags & RENAME_EXCHANGE))
4344 error = may_delete(new_dir, new_dentry, is_dir);
4346 error = may_delete(new_dir, new_dentry, new_is_dir);
4351 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4354 if (flags && !old_dir->i_op->rename2)
4358 * If we are going to change the parent - check write permissions,
4359 * we'll need to flip '..'.
4361 if (new_dir != old_dir) {
4363 error = inode_permission(source, MAY_WRITE);
4367 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4368 error = inode_permission(target, MAY_WRITE);
4374 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4379 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4381 if (!is_dir || (flags & RENAME_EXCHANGE))
4382 lock_two_nondirectories(source, target);
4387 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4390 if (max_links && new_dir != old_dir) {
4392 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4394 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4395 old_dir->i_nlink >= max_links)
4398 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4399 shrink_dcache_parent(new_dentry);
4401 error = try_break_deleg(source, delegated_inode);
4405 if (target && !new_is_dir) {
4406 error = try_break_deleg(target, delegated_inode);
4410 if (!old_dir->i_op->rename2) {
4411 error = old_dir->i_op->rename(old_dir, old_dentry,
4412 new_dir, new_dentry);
4414 WARN_ON(old_dir->i_op->rename != NULL);
4415 error = old_dir->i_op->rename2(old_dir, old_dentry,
4416 new_dir, new_dentry, flags);
4421 if (!(flags & RENAME_EXCHANGE) && target) {
4423 target->i_flags |= S_DEAD;
4424 dont_mount(new_dentry);
4425 detach_mounts(new_dentry);
4427 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4428 if (!(flags & RENAME_EXCHANGE))
4429 d_move(old_dentry, new_dentry);
4431 d_exchange(old_dentry, new_dentry);
4434 if (!is_dir || (flags & RENAME_EXCHANGE))
4435 unlock_two_nondirectories(source, target);
4437 inode_unlock(target);
4440 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4441 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4442 if (flags & RENAME_EXCHANGE) {
4443 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4444 new_is_dir, NULL, new_dentry);
4447 fsnotify_oldname_free(old_name);
4451 EXPORT_SYMBOL(vfs_rename);
4453 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4454 int, newdfd, const char __user *, newname, unsigned int, flags)
4456 struct dentry *old_dentry, *new_dentry;
4457 struct dentry *trap;
4458 struct path old_path, new_path;
4459 struct qstr old_last, new_last;
4460 int old_type, new_type;
4461 struct inode *delegated_inode = NULL;
4462 struct filename *from;
4463 struct filename *to;
4464 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4465 bool should_retry = false;
4468 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4471 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4472 (flags & RENAME_EXCHANGE))
4475 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4478 if (flags & RENAME_EXCHANGE)
4482 from = user_path_parent(olddfd, oldname,
4483 &old_path, &old_last, &old_type, lookup_flags);
4485 error = PTR_ERR(from);
4489 to = user_path_parent(newdfd, newname,
4490 &new_path, &new_last, &new_type, lookup_flags);
4492 error = PTR_ERR(to);
4497 if (old_path.mnt != new_path.mnt)
4501 if (old_type != LAST_NORM)
4504 if (flags & RENAME_NOREPLACE)
4506 if (new_type != LAST_NORM)
4509 error = mnt_want_write(old_path.mnt);
4514 trap = lock_rename(new_path.dentry, old_path.dentry);
4516 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4517 error = PTR_ERR(old_dentry);
4518 if (IS_ERR(old_dentry))
4520 /* source must exist */
4522 if (d_is_negative(old_dentry))
4524 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4525 error = PTR_ERR(new_dentry);
4526 if (IS_ERR(new_dentry))
4529 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4531 if (flags & RENAME_EXCHANGE) {
4533 if (d_is_negative(new_dentry))
4536 if (!d_is_dir(new_dentry)) {
4538 if (new_last.name[new_last.len])
4542 /* unless the source is a directory trailing slashes give -ENOTDIR */
4543 if (!d_is_dir(old_dentry)) {
4545 if (old_last.name[old_last.len])
4547 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4550 /* source should not be ancestor of target */
4552 if (old_dentry == trap)
4554 /* target should not be an ancestor of source */
4555 if (!(flags & RENAME_EXCHANGE))
4557 if (new_dentry == trap)
4560 error = security_path_rename(&old_path, old_dentry,
4561 &new_path, new_dentry, flags);
4564 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4565 new_path.dentry->d_inode, new_dentry,
4566 &delegated_inode, flags);
4572 unlock_rename(new_path.dentry, old_path.dentry);
4573 if (delegated_inode) {
4574 error = break_deleg_wait(&delegated_inode);
4578 mnt_drop_write(old_path.mnt);
4580 if (retry_estale(error, lookup_flags))
4581 should_retry = true;
4582 path_put(&new_path);
4585 path_put(&old_path);
4588 should_retry = false;
4589 lookup_flags |= LOOKUP_REVAL;
4596 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4597 int, newdfd, const char __user *, newname)
4599 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4602 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4604 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4607 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4609 int error = may_create(dir, dentry);
4613 if (!dir->i_op->mknod)
4616 return dir->i_op->mknod(dir, dentry,
4617 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4619 EXPORT_SYMBOL(vfs_whiteout);
4621 int readlink_copy(char __user *buffer, int buflen, const char *link)
4623 int len = PTR_ERR(link);
4628 if (len > (unsigned) buflen)
4630 if (copy_to_user(buffer, link, len))
4637 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4638 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4639 * for any given inode is up to filesystem.
4641 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4643 DEFINE_DELAYED_CALL(done);
4644 struct inode *inode = d_inode(dentry);
4645 const char *link = inode->i_link;
4649 link = inode->i_op->get_link(dentry, inode, &done);
4651 return PTR_ERR(link);
4653 res = readlink_copy(buffer, buflen, link);
4654 do_delayed_call(&done);
4657 EXPORT_SYMBOL(generic_readlink);
4659 /* get the link contents into pagecache */
4660 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4661 struct delayed_call *callback)
4665 struct address_space *mapping = inode->i_mapping;
4668 page = find_get_page(mapping, 0);
4670 return ERR_PTR(-ECHILD);
4671 if (!PageUptodate(page)) {
4673 return ERR_PTR(-ECHILD);
4676 page = read_mapping_page(mapping, 0, NULL);
4680 set_delayed_call(callback, page_put_link, page);
4681 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4682 kaddr = page_address(page);
4683 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4687 EXPORT_SYMBOL(page_get_link);
4689 void page_put_link(void *arg)
4693 EXPORT_SYMBOL(page_put_link);
4695 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4697 DEFINE_DELAYED_CALL(done);
4698 int res = readlink_copy(buffer, buflen,
4699 page_get_link(dentry, d_inode(dentry),
4701 do_delayed_call(&done);
4704 EXPORT_SYMBOL(page_readlink);
4707 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4709 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4711 struct address_space *mapping = inode->i_mapping;
4715 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4717 flags |= AOP_FLAG_NOFS;
4720 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4721 flags, &page, &fsdata);
4725 memcpy(page_address(page), symname, len-1);
4727 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4734 mark_inode_dirty(inode);
4739 EXPORT_SYMBOL(__page_symlink);
4741 int page_symlink(struct inode *inode, const char *symname, int len)
4743 return __page_symlink(inode, symname, len,
4744 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4746 EXPORT_SYMBOL(page_symlink);
4748 const struct inode_operations page_symlink_inode_operations = {
4749 .readlink = generic_readlink,
4750 .get_link = page_get_link,
4752 EXPORT_SYMBOL(page_symlink_inode_operations);