4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <linux/bitops.h>
39 #include <linux/init_task.h>
40 #include <linux/uaccess.h>
45 /* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
73 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
79 * the name is a symlink pointing to a non-existent name.
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
90 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
93 * [10-Sep-98 Alan Modra] Another symlink change.
96 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
113 * any extra contention...
116 /* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
124 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127 getname_flags(const char __user *filename, int flags, int *empty)
129 struct filename *result;
133 result = audit_reusename(filename);
137 result = __getname();
138 if (unlikely(!result))
139 return ERR_PTR(-ENOMEM);
142 * First, try to embed the struct filename inside the names_cache
145 kname = (char *)result->iname;
146 result->name = kname;
148 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
149 if (unlikely(len < 0)) {
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
160 if (unlikely(len == EMBEDDED_NAME_MAX)) {
161 const size_t size = offsetof(struct filename, iname[1]);
162 kname = (char *)result;
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
169 result = kzalloc(size, GFP_KERNEL);
170 if (unlikely(!result)) {
172 return ERR_PTR(-ENOMEM);
174 result->name = kname;
175 len = strncpy_from_user(kname, filename, PATH_MAX);
176 if (unlikely(len < 0)) {
181 if (unlikely(len == PATH_MAX)) {
184 return ERR_PTR(-ENAMETOOLONG);
189 /* The empty path is special. */
190 if (unlikely(!len)) {
193 if (!(flags & LOOKUP_EMPTY)) {
195 return ERR_PTR(-ENOENT);
199 result->uptr = filename;
200 result->aname = NULL;
201 audit_getname(result);
206 getname(const char __user * filename)
208 return getname_flags(filename, 0, NULL);
212 getname_kernel(const char * filename)
214 struct filename *result;
215 int len = strlen(filename) + 1;
217 result = __getname();
218 if (unlikely(!result))
219 return ERR_PTR(-ENOMEM);
221 if (len <= EMBEDDED_NAME_MAX) {
222 result->name = (char *)result->iname;
223 } else if (len <= PATH_MAX) {
224 struct filename *tmp;
226 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 if (unlikely(!tmp)) {
229 return ERR_PTR(-ENOMEM);
231 tmp->name = (char *)result;
235 return ERR_PTR(-ENAMETOOLONG);
237 memcpy((char *)result->name, filename, len);
239 result->aname = NULL;
241 audit_getname(result);
246 void putname(struct filename *name)
248 BUG_ON(name->refcnt <= 0);
250 if (--name->refcnt > 0)
253 if (name->name != name->iname) {
254 __putname(name->name);
260 static int check_acl(struct inode *inode, int mask)
262 #ifdef CONFIG_FS_POSIX_ACL
263 struct posix_acl *acl;
265 if (mask & MAY_NOT_BLOCK) {
266 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
269 /* no ->get_acl() calls in RCU mode... */
270 if (is_uncached_acl(acl))
272 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
275 acl = get_acl(inode, ACL_TYPE_ACCESS);
279 int error = posix_acl_permission(inode, acl, mask);
280 posix_acl_release(acl);
289 * This does the basic permission checking
291 static int acl_permission_check(struct inode *inode, int mask)
293 unsigned int mode = inode->i_mode;
295 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
298 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
299 int error = check_acl(inode, mask);
300 if (error != -EAGAIN)
304 if (in_group_p(inode->i_gid))
309 * If the DACs are ok we don't need any capability check.
311 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
317 * generic_permission - check for access rights on a Posix-like filesystem
318 * @inode: inode to check access rights for
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
330 int generic_permission(struct inode *inode, int mask)
335 * Do the basic permission checks.
337 ret = acl_permission_check(inode, mask);
341 if (S_ISDIR(inode->i_mode)) {
342 /* DACs are overridable for directories */
343 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
352 * Read/write DACs are always overridable.
353 * Executable DACs are overridable when there is
354 * at least one exec bit set.
356 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
361 * Searching includes executable on directories, else just read.
363 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
364 if (mask == MAY_READ)
365 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
370 EXPORT_SYMBOL(generic_permission);
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
378 static inline int do_inode_permission(struct inode *inode, int mask)
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
389 return generic_permission(inode, mask);
393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
397 * Check for read/write/execute permissions on an inode.
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
404 int __inode_permission(struct inode *inode, int mask)
408 if (unlikely(mask & MAY_WRITE)) {
410 * Nobody gets write access to an immutable file.
412 if (IS_IMMUTABLE(inode))
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
420 if (HAS_UNMAPPED_ID(inode))
424 retval = do_inode_permission(inode, mask);
428 retval = devcgroup_inode_permission(inode, mask);
432 return security_inode_permission(inode, mask);
434 EXPORT_SYMBOL(__inode_permission);
437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
439 * @inode: Inode to check permission on
440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
442 * Separate out file-system wide checks from inode-specific permission checks.
444 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
446 if (unlikely(mask & MAY_WRITE)) {
447 umode_t mode = inode->i_mode;
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb->s_flags & MS_RDONLY) &&
451 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
468 int inode_permission(struct inode *inode, int mask)
472 retval = sb_permission(inode->i_sb, inode, mask);
475 return __inode_permission(inode, mask);
477 EXPORT_SYMBOL(inode_permission);
480 * path_get - get a reference to a path
481 * @path: path to get the reference to
483 * Given a path increment the reference count to the dentry and the vfsmount.
485 void path_get(const struct path *path)
490 EXPORT_SYMBOL(path_get);
493 * path_put - put a reference to a path
494 * @path: path to put the reference to
496 * Given a path decrement the reference count to the dentry and the vfsmount.
498 void path_put(const struct path *path)
503 EXPORT_SYMBOL(path_put);
505 #define EMBEDDED_LEVELS 2
510 struct inode *inode; /* path.dentry.d_inode */
515 int total_link_count;
518 struct delayed_call done;
521 } *stack, internal[EMBEDDED_LEVELS];
522 struct filename *name;
523 struct nameidata *saved;
524 struct inode *link_inode;
529 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
535 p->total_link_count = old ? old->total_link_count : 0;
537 current->nameidata = p;
540 static void restore_nameidata(void)
542 struct nameidata *now = current->nameidata, *old = now->saved;
544 current->nameidata = old;
546 old->total_link_count = now->total_link_count;
547 if (now->stack != now->internal)
551 static int __nd_alloc_stack(struct nameidata *nd)
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
566 memcpy(p, nd->internal, sizeof(nd->internal));
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
578 static bool path_connected(const struct path *path)
580 struct vfsmount *mnt = path->mnt;
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
586 return is_subdir(path->dentry, mnt->mnt_root);
589 static inline int nd_alloc_stack(struct nameidata *nd)
591 if (likely(nd->depth != EMBEDDED_LEVELS))
593 if (likely(nd->stack != nd->internal))
595 return __nd_alloc_stack(nd);
598 static void drop_links(struct nameidata *nd)
602 struct saved *last = nd->stack + i;
603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
608 static void terminate_walk(struct nameidata *nd)
611 if (!(nd->flags & LOOKUP_RCU)) {
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
629 /* path_put is needed afterwards regardless of success or failure */
630 static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
647 static bool legitimize_links(struct nameidata *nd)
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
662 * Path walking has 2 modes, rcu-walk and ref-walk (see
663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * Returns: 0 on success, -ECHILD on failure
677 * unlazy_walk attempts to legitimize the current nd->path and nd->root
679 * Must be called from rcu-walk context.
680 * Nothing should touch nameidata between unlazy_walk() failure and
683 static int unlazy_walk(struct nameidata *nd)
685 struct dentry *parent = nd->path.dentry;
687 BUG_ON(!(nd->flags & LOOKUP_RCU));
689 nd->flags &= ~LOOKUP_RCU;
690 if (unlikely(!legitimize_links(nd)))
692 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
694 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
695 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
699 BUG_ON(nd->inode != parent->d_inode);
704 nd->path.dentry = NULL;
706 if (!(nd->flags & LOOKUP_ROOT))
714 * unlazy_child - try to switch to ref-walk mode.
715 * @nd: nameidata pathwalk data
716 * @dentry: child of nd->path.dentry
717 * @seq: seq number to check dentry against
718 * Returns: 0 on success, -ECHILD on failure
720 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
721 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
722 * @nd. Must be called from rcu-walk context.
723 * Nothing should touch nameidata between unlazy_child() failure and
726 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
728 BUG_ON(!(nd->flags & LOOKUP_RCU));
730 nd->flags &= ~LOOKUP_RCU;
731 if (unlikely(!legitimize_links(nd)))
733 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
735 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
739 * We need to move both the parent and the dentry from the RCU domain
740 * to be properly refcounted. And the sequence number in the dentry
741 * validates *both* dentry counters, since we checked the sequence
742 * number of the parent after we got the child sequence number. So we
743 * know the parent must still be valid if the child sequence number is
745 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
747 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
753 * Sequence counts matched. Now make sure that the root is
754 * still valid and get it if required.
756 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
757 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
770 nd->path.dentry = NULL;
774 if (!(nd->flags & LOOKUP_ROOT))
779 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
781 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
782 return dentry->d_op->d_revalidate(dentry, flags);
788 * complete_walk - successful completion of path walk
789 * @nd: pointer nameidata
791 * If we had been in RCU mode, drop out of it and legitimize nd->path.
792 * Revalidate the final result, unless we'd already done that during
793 * the path walk or the filesystem doesn't ask for it. Return 0 on
794 * success, -error on failure. In case of failure caller does not
795 * need to drop nd->path.
797 static int complete_walk(struct nameidata *nd)
799 struct dentry *dentry = nd->path.dentry;
802 if (nd->flags & LOOKUP_RCU) {
803 if (!(nd->flags & LOOKUP_ROOT))
805 if (unlikely(unlazy_walk(nd)))
809 if (likely(!(nd->flags & LOOKUP_JUMPED)))
812 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
815 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
825 static void set_root(struct nameidata *nd)
827 struct fs_struct *fs = current->fs;
829 if (nd->flags & LOOKUP_RCU) {
833 seq = read_seqcount_begin(&fs->seq);
835 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
836 } while (read_seqcount_retry(&fs->seq, seq));
838 get_fs_root(fs, &nd->root);
842 static void path_put_conditional(struct path *path, struct nameidata *nd)
845 if (path->mnt != nd->path.mnt)
849 static inline void path_to_nameidata(const struct path *path,
850 struct nameidata *nd)
852 if (!(nd->flags & LOOKUP_RCU)) {
853 dput(nd->path.dentry);
854 if (nd->path.mnt != path->mnt)
855 mntput(nd->path.mnt);
857 nd->path.mnt = path->mnt;
858 nd->path.dentry = path->dentry;
861 static int nd_jump_root(struct nameidata *nd)
863 if (nd->flags & LOOKUP_RCU) {
867 nd->inode = d->d_inode;
868 nd->seq = nd->root_seq;
869 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
875 nd->inode = nd->path.dentry->d_inode;
877 nd->flags |= LOOKUP_JUMPED;
882 * Helper to directly jump to a known parsed path from ->get_link,
883 * caller must have taken a reference to path beforehand.
885 void nd_jump_link(struct path *path)
887 struct nameidata *nd = current->nameidata;
891 nd->inode = nd->path.dentry->d_inode;
892 nd->flags |= LOOKUP_JUMPED;
895 static inline void put_link(struct nameidata *nd)
897 struct saved *last = nd->stack + --nd->depth;
898 do_delayed_call(&last->done);
899 if (!(nd->flags & LOOKUP_RCU))
900 path_put(&last->link);
903 int sysctl_protected_symlinks __read_mostly = 0;
904 int sysctl_protected_hardlinks __read_mostly = 0;
907 * may_follow_link - Check symlink following for unsafe situations
908 * @nd: nameidata pathwalk data
910 * In the case of the sysctl_protected_symlinks sysctl being enabled,
911 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
912 * in a sticky world-writable directory. This is to protect privileged
913 * processes from failing races against path names that may change out
914 * from under them by way of other users creating malicious symlinks.
915 * It will permit symlinks to be followed only when outside a sticky
916 * world-writable directory, or when the uid of the symlink and follower
917 * match, or when the directory owner matches the symlink's owner.
919 * Returns 0 if following the symlink is allowed, -ve on error.
921 static inline int may_follow_link(struct nameidata *nd)
923 const struct inode *inode;
924 const struct inode *parent;
927 if (!sysctl_protected_symlinks)
930 /* Allowed if owner and follower match. */
931 inode = nd->link_inode;
932 if (uid_eq(current_cred()->fsuid, inode->i_uid))
935 /* Allowed if parent directory not sticky and world-writable. */
937 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
940 /* Allowed if parent directory and link owner match. */
941 puid = parent->i_uid;
942 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
945 if (nd->flags & LOOKUP_RCU)
948 audit_log_link_denied("follow_link", &nd->stack[0].link);
953 * safe_hardlink_source - Check for safe hardlink conditions
954 * @inode: the source inode to hardlink from
956 * Return false if at least one of the following conditions:
957 * - inode is not a regular file
959 * - inode is setgid and group-exec
960 * - access failure for read and write
962 * Otherwise returns true.
964 static bool safe_hardlink_source(struct inode *inode)
966 umode_t mode = inode->i_mode;
968 /* Special files should not get pinned to the filesystem. */
972 /* Setuid files should not get pinned to the filesystem. */
976 /* Executable setgid files should not get pinned to the filesystem. */
977 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
980 /* Hardlinking to unreadable or unwritable sources is dangerous. */
981 if (inode_permission(inode, MAY_READ | MAY_WRITE))
988 * may_linkat - Check permissions for creating a hardlink
989 * @link: the source to hardlink from
991 * Block hardlink when all of:
992 * - sysctl_protected_hardlinks enabled
993 * - fsuid does not match inode
994 * - hardlink source is unsafe (see safe_hardlink_source() above)
995 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
997 * Returns 0 if successful, -ve on error.
999 static int may_linkat(struct path *link)
1001 struct inode *inode;
1003 if (!sysctl_protected_hardlinks)
1006 inode = link->dentry->d_inode;
1008 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1009 * otherwise, it must be a safe source.
1011 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
1014 audit_log_link_denied("linkat", link);
1018 static __always_inline
1019 const char *get_link(struct nameidata *nd)
1021 struct saved *last = nd->stack + nd->depth - 1;
1022 struct dentry *dentry = last->link.dentry;
1023 struct inode *inode = nd->link_inode;
1027 if (!(nd->flags & LOOKUP_RCU)) {
1028 touch_atime(&last->link);
1030 } else if (atime_needs_update_rcu(&last->link, inode)) {
1031 if (unlikely(unlazy_walk(nd)))
1032 return ERR_PTR(-ECHILD);
1033 touch_atime(&last->link);
1036 error = security_inode_follow_link(dentry, inode,
1037 nd->flags & LOOKUP_RCU);
1038 if (unlikely(error))
1039 return ERR_PTR(error);
1041 nd->last_type = LAST_BIND;
1042 res = inode->i_link;
1044 const char * (*get)(struct dentry *, struct inode *,
1045 struct delayed_call *);
1046 get = inode->i_op->get_link;
1047 if (nd->flags & LOOKUP_RCU) {
1048 res = get(NULL, inode, &last->done);
1049 if (res == ERR_PTR(-ECHILD)) {
1050 if (unlikely(unlazy_walk(nd)))
1051 return ERR_PTR(-ECHILD);
1052 res = get(dentry, inode, &last->done);
1055 res = get(dentry, inode, &last->done);
1057 if (IS_ERR_OR_NULL(res))
1063 if (unlikely(nd_jump_root(nd)))
1064 return ERR_PTR(-ECHILD);
1065 while (unlikely(*++res == '/'))
1074 * follow_up - Find the mountpoint of path's vfsmount
1076 * Given a path, find the mountpoint of its source file system.
1077 * Replace @path with the path of the mountpoint in the parent mount.
1080 * Return 1 if we went up a level and 0 if we were already at the
1083 int follow_up(struct path *path)
1085 struct mount *mnt = real_mount(path->mnt);
1086 struct mount *parent;
1087 struct dentry *mountpoint;
1089 read_seqlock_excl(&mount_lock);
1090 parent = mnt->mnt_parent;
1091 if (parent == mnt) {
1092 read_sequnlock_excl(&mount_lock);
1095 mntget(&parent->mnt);
1096 mountpoint = dget(mnt->mnt_mountpoint);
1097 read_sequnlock_excl(&mount_lock);
1099 path->dentry = mountpoint;
1101 path->mnt = &parent->mnt;
1104 EXPORT_SYMBOL(follow_up);
1107 * Perform an automount
1108 * - return -EISDIR to tell follow_managed() to stop and return the path we
1111 static int follow_automount(struct path *path, struct nameidata *nd,
1114 struct vfsmount *mnt;
1117 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1120 /* We don't want to mount if someone's just doing a stat -
1121 * unless they're stat'ing a directory and appended a '/' to
1124 * We do, however, want to mount if someone wants to open or
1125 * create a file of any type under the mountpoint, wants to
1126 * traverse through the mountpoint or wants to open the
1127 * mounted directory. Also, autofs may mark negative dentries
1128 * as being automount points. These will need the attentions
1129 * of the daemon to instantiate them before they can be used.
1131 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1132 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1133 path->dentry->d_inode)
1136 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1139 nd->total_link_count++;
1140 if (nd->total_link_count >= 40)
1143 mnt = path->dentry->d_op->d_automount(path);
1146 * The filesystem is allowed to return -EISDIR here to indicate
1147 * it doesn't want to automount. For instance, autofs would do
1148 * this so that its userspace daemon can mount on this dentry.
1150 * However, we can only permit this if it's a terminal point in
1151 * the path being looked up; if it wasn't then the remainder of
1152 * the path is inaccessible and we should say so.
1154 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1156 return PTR_ERR(mnt);
1159 if (!mnt) /* mount collision */
1162 if (!*need_mntput) {
1163 /* lock_mount() may release path->mnt on error */
1165 *need_mntput = true;
1167 err = finish_automount(mnt, path);
1171 /* Someone else made a mount here whilst we were busy */
1176 path->dentry = dget(mnt->mnt_root);
1185 * Handle a dentry that is managed in some way.
1186 * - Flagged for transit management (autofs)
1187 * - Flagged as mountpoint
1188 * - Flagged as automount point
1190 * This may only be called in refwalk mode.
1192 * Serialization is taken care of in namespace.c
1194 static int follow_managed(struct path *path, struct nameidata *nd)
1196 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1198 bool need_mntput = false;
1201 /* Given that we're not holding a lock here, we retain the value in a
1202 * local variable for each dentry as we look at it so that we don't see
1203 * the components of that value change under us */
1204 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1205 managed &= DCACHE_MANAGED_DENTRY,
1206 unlikely(managed != 0)) {
1207 /* Allow the filesystem to manage the transit without i_mutex
1209 if (managed & DCACHE_MANAGE_TRANSIT) {
1210 BUG_ON(!path->dentry->d_op);
1211 BUG_ON(!path->dentry->d_op->d_manage);
1212 ret = path->dentry->d_op->d_manage(path, false);
1217 /* Transit to a mounted filesystem. */
1218 if (managed & DCACHE_MOUNTED) {
1219 struct vfsmount *mounted = lookup_mnt(path);
1224 path->mnt = mounted;
1225 path->dentry = dget(mounted->mnt_root);
1230 /* Something is mounted on this dentry in another
1231 * namespace and/or whatever was mounted there in this
1232 * namespace got unmounted before lookup_mnt() could
1236 /* Handle an automount point */
1237 if (managed & DCACHE_NEED_AUTOMOUNT) {
1238 ret = follow_automount(path, nd, &need_mntput);
1244 /* We didn't change the current path point */
1248 if (need_mntput && path->mnt == mnt)
1250 if (ret == -EISDIR || !ret)
1253 nd->flags |= LOOKUP_JUMPED;
1254 if (unlikely(ret < 0))
1255 path_put_conditional(path, nd);
1259 int follow_down_one(struct path *path)
1261 struct vfsmount *mounted;
1263 mounted = lookup_mnt(path);
1267 path->mnt = mounted;
1268 path->dentry = dget(mounted->mnt_root);
1273 EXPORT_SYMBOL(follow_down_one);
1275 static inline int managed_dentry_rcu(const struct path *path)
1277 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1278 path->dentry->d_op->d_manage(path, true) : 0;
1282 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1283 * we meet a managed dentry that would need blocking.
1285 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1286 struct inode **inode, unsigned *seqp)
1289 struct mount *mounted;
1291 * Don't forget we might have a non-mountpoint managed dentry
1292 * that wants to block transit.
1294 switch (managed_dentry_rcu(path)) {
1304 if (!d_mountpoint(path->dentry))
1305 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1307 mounted = __lookup_mnt(path->mnt, path->dentry);
1310 path->mnt = &mounted->mnt;
1311 path->dentry = mounted->mnt.mnt_root;
1312 nd->flags |= LOOKUP_JUMPED;
1313 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1315 * Update the inode too. We don't need to re-check the
1316 * dentry sequence number here after this d_inode read,
1317 * because a mount-point is always pinned.
1319 *inode = path->dentry->d_inode;
1321 return !read_seqretry(&mount_lock, nd->m_seq) &&
1322 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1325 static int follow_dotdot_rcu(struct nameidata *nd)
1327 struct inode *inode = nd->inode;
1330 if (path_equal(&nd->path, &nd->root))
1332 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1333 struct dentry *old = nd->path.dentry;
1334 struct dentry *parent = old->d_parent;
1337 inode = parent->d_inode;
1338 seq = read_seqcount_begin(&parent->d_seq);
1339 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1341 nd->path.dentry = parent;
1343 if (unlikely(!path_connected(&nd->path)))
1347 struct mount *mnt = real_mount(nd->path.mnt);
1348 struct mount *mparent = mnt->mnt_parent;
1349 struct dentry *mountpoint = mnt->mnt_mountpoint;
1350 struct inode *inode2 = mountpoint->d_inode;
1351 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1352 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1354 if (&mparent->mnt == nd->path.mnt)
1356 /* we know that mountpoint was pinned */
1357 nd->path.dentry = mountpoint;
1358 nd->path.mnt = &mparent->mnt;
1363 while (unlikely(d_mountpoint(nd->path.dentry))) {
1364 struct mount *mounted;
1365 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1366 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1370 nd->path.mnt = &mounted->mnt;
1371 nd->path.dentry = mounted->mnt.mnt_root;
1372 inode = nd->path.dentry->d_inode;
1373 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1380 * Follow down to the covering mount currently visible to userspace. At each
1381 * point, the filesystem owning that dentry may be queried as to whether the
1382 * caller is permitted to proceed or not.
1384 int follow_down(struct path *path)
1389 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1390 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1391 /* Allow the filesystem to manage the transit without i_mutex
1394 * We indicate to the filesystem if someone is trying to mount
1395 * something here. This gives autofs the chance to deny anyone
1396 * other than its daemon the right to mount on its
1399 * The filesystem may sleep at this point.
1401 if (managed & DCACHE_MANAGE_TRANSIT) {
1402 BUG_ON(!path->dentry->d_op);
1403 BUG_ON(!path->dentry->d_op->d_manage);
1404 ret = path->dentry->d_op->d_manage(path, false);
1406 return ret == -EISDIR ? 0 : ret;
1409 /* Transit to a mounted filesystem. */
1410 if (managed & DCACHE_MOUNTED) {
1411 struct vfsmount *mounted = lookup_mnt(path);
1416 path->mnt = mounted;
1417 path->dentry = dget(mounted->mnt_root);
1421 /* Don't handle automount points here */
1426 EXPORT_SYMBOL(follow_down);
1429 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1431 static void follow_mount(struct path *path)
1433 while (d_mountpoint(path->dentry)) {
1434 struct vfsmount *mounted = lookup_mnt(path);
1439 path->mnt = mounted;
1440 path->dentry = dget(mounted->mnt_root);
1444 static int path_parent_directory(struct path *path)
1446 struct dentry *old = path->dentry;
1447 /* rare case of legitimate dget_parent()... */
1448 path->dentry = dget_parent(path->dentry);
1450 if (unlikely(!path_connected(path)))
1455 static int follow_dotdot(struct nameidata *nd)
1458 if (nd->path.dentry == nd->root.dentry &&
1459 nd->path.mnt == nd->root.mnt) {
1462 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1463 int ret = path_parent_directory(&nd->path);
1468 if (!follow_up(&nd->path))
1471 follow_mount(&nd->path);
1472 nd->inode = nd->path.dentry->d_inode;
1477 * This looks up the name in dcache and possibly revalidates the found dentry.
1478 * NULL is returned if the dentry does not exist in the cache.
1480 static struct dentry *lookup_dcache(const struct qstr *name,
1484 struct dentry *dentry = d_lookup(dir, name);
1486 int error = d_revalidate(dentry, flags);
1487 if (unlikely(error <= 0)) {
1489 d_invalidate(dentry);
1491 return ERR_PTR(error);
1498 * Call i_op->lookup on the dentry. The dentry must be negative and
1501 * dir->d_inode->i_mutex must be held
1503 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1508 /* Don't create child dentry for a dead directory. */
1509 if (unlikely(IS_DEADDIR(dir))) {
1511 return ERR_PTR(-ENOENT);
1514 old = dir->i_op->lookup(dir, dentry, flags);
1515 if (unlikely(old)) {
1522 static struct dentry *__lookup_hash(const struct qstr *name,
1523 struct dentry *base, unsigned int flags)
1525 struct dentry *dentry = lookup_dcache(name, base, flags);
1530 dentry = d_alloc(base, name);
1531 if (unlikely(!dentry))
1532 return ERR_PTR(-ENOMEM);
1534 return lookup_real(base->d_inode, dentry, flags);
1537 static int lookup_fast(struct nameidata *nd,
1538 struct path *path, struct inode **inode,
1541 struct vfsmount *mnt = nd->path.mnt;
1542 struct dentry *dentry, *parent = nd->path.dentry;
1547 * Rename seqlock is not required here because in the off chance
1548 * of a false negative due to a concurrent rename, the caller is
1549 * going to fall back to non-racy lookup.
1551 if (nd->flags & LOOKUP_RCU) {
1554 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1555 if (unlikely(!dentry)) {
1556 if (unlazy_walk(nd))
1562 * This sequence count validates that the inode matches
1563 * the dentry name information from lookup.
1565 *inode = d_backing_inode(dentry);
1566 negative = d_is_negative(dentry);
1567 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1571 * This sequence count validates that the parent had no
1572 * changes while we did the lookup of the dentry above.
1574 * The memory barrier in read_seqcount_begin of child is
1575 * enough, we can use __read_seqcount_retry here.
1577 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1581 status = d_revalidate(dentry, nd->flags);
1582 if (likely(status > 0)) {
1584 * Note: do negative dentry check after revalidation in
1585 * case that drops it.
1587 if (unlikely(negative))
1590 path->dentry = dentry;
1591 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1594 if (unlazy_child(nd, dentry, seq))
1596 if (unlikely(status == -ECHILD))
1597 /* we'd been told to redo it in non-rcu mode */
1598 status = d_revalidate(dentry, nd->flags);
1600 dentry = __d_lookup(parent, &nd->last);
1601 if (unlikely(!dentry))
1603 status = d_revalidate(dentry, nd->flags);
1605 if (unlikely(status <= 0)) {
1607 d_invalidate(dentry);
1611 if (unlikely(d_is_negative(dentry))) {
1617 path->dentry = dentry;
1618 err = follow_managed(path, nd);
1619 if (likely(err > 0))
1620 *inode = d_backing_inode(path->dentry);
1624 /* Fast lookup failed, do it the slow way */
1625 static struct dentry *lookup_slow(const struct qstr *name,
1629 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1630 struct inode *inode = dir->d_inode;
1631 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1633 inode_lock_shared(inode);
1634 /* Don't go there if it's already dead */
1635 if (unlikely(IS_DEADDIR(inode)))
1638 dentry = d_alloc_parallel(dir, name, &wq);
1641 if (unlikely(!d_in_lookup(dentry))) {
1642 if (!(flags & LOOKUP_NO_REVAL)) {
1643 int error = d_revalidate(dentry, flags);
1644 if (unlikely(error <= 0)) {
1646 d_invalidate(dentry);
1651 dentry = ERR_PTR(error);
1655 old = inode->i_op->lookup(inode, dentry, flags);
1656 d_lookup_done(dentry);
1657 if (unlikely(old)) {
1663 inode_unlock_shared(inode);
1667 static inline int may_lookup(struct nameidata *nd)
1669 if (nd->flags & LOOKUP_RCU) {
1670 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1673 if (unlazy_walk(nd))
1676 return inode_permission(nd->inode, MAY_EXEC);
1679 static inline int handle_dots(struct nameidata *nd, int type)
1681 if (type == LAST_DOTDOT) {
1684 if (nd->flags & LOOKUP_RCU) {
1685 return follow_dotdot_rcu(nd);
1687 return follow_dotdot(nd);
1692 static int pick_link(struct nameidata *nd, struct path *link,
1693 struct inode *inode, unsigned seq)
1697 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1698 path_to_nameidata(link, nd);
1701 if (!(nd->flags & LOOKUP_RCU)) {
1702 if (link->mnt == nd->path.mnt)
1705 error = nd_alloc_stack(nd);
1706 if (unlikely(error)) {
1707 if (error == -ECHILD) {
1708 if (unlikely(!legitimize_path(nd, link, seq))) {
1711 nd->flags &= ~LOOKUP_RCU;
1712 nd->path.mnt = NULL;
1713 nd->path.dentry = NULL;
1714 if (!(nd->flags & LOOKUP_ROOT))
1715 nd->root.mnt = NULL;
1717 } else if (likely(unlazy_walk(nd)) == 0)
1718 error = nd_alloc_stack(nd);
1726 last = nd->stack + nd->depth++;
1728 clear_delayed_call(&last->done);
1729 nd->link_inode = inode;
1734 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1737 * Do we need to follow links? We _really_ want to be able
1738 * to do this check without having to look at inode->i_op,
1739 * so we keep a cache of "no, this doesn't need follow_link"
1740 * for the common case.
1742 static inline int step_into(struct nameidata *nd, struct path *path,
1743 int flags, struct inode *inode, unsigned seq)
1745 if (!(flags & WALK_MORE) && nd->depth)
1747 if (likely(!d_is_symlink(path->dentry)) ||
1748 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1749 /* not a symlink or should not follow */
1750 path_to_nameidata(path, nd);
1755 /* make sure that d_is_symlink above matches inode */
1756 if (nd->flags & LOOKUP_RCU) {
1757 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1760 return pick_link(nd, path, inode, seq);
1763 static int walk_component(struct nameidata *nd, int flags)
1766 struct inode *inode;
1770 * "." and ".." are special - ".." especially so because it has
1771 * to be able to know about the current root directory and
1772 * parent relationships.
1774 if (unlikely(nd->last_type != LAST_NORM)) {
1775 err = handle_dots(nd, nd->last_type);
1776 if (!(flags & WALK_MORE) && nd->depth)
1780 err = lookup_fast(nd, &path, &inode, &seq);
1781 if (unlikely(err <= 0)) {
1784 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1786 if (IS_ERR(path.dentry))
1787 return PTR_ERR(path.dentry);
1789 path.mnt = nd->path.mnt;
1790 err = follow_managed(&path, nd);
1791 if (unlikely(err < 0))
1794 if (unlikely(d_is_negative(path.dentry))) {
1795 path_to_nameidata(&path, nd);
1799 seq = 0; /* we are already out of RCU mode */
1800 inode = d_backing_inode(path.dentry);
1803 return step_into(nd, &path, flags, inode, seq);
1807 * We can do the critical dentry name comparison and hashing
1808 * operations one word at a time, but we are limited to:
1810 * - Architectures with fast unaligned word accesses. We could
1811 * do a "get_unaligned()" if this helps and is sufficiently
1814 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1815 * do not trap on the (extremely unlikely) case of a page
1816 * crossing operation.
1818 * - Furthermore, we need an efficient 64-bit compile for the
1819 * 64-bit case in order to generate the "number of bytes in
1820 * the final mask". Again, that could be replaced with a
1821 * efficient population count instruction or similar.
1823 #ifdef CONFIG_DCACHE_WORD_ACCESS
1825 #include <asm/word-at-a-time.h>
1829 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1831 #elif defined(CONFIG_64BIT)
1833 * Register pressure in the mixing function is an issue, particularly
1834 * on 32-bit x86, but almost any function requires one state value and
1835 * one temporary. Instead, use a function designed for two state values
1836 * and no temporaries.
1838 * This function cannot create a collision in only two iterations, so
1839 * we have two iterations to achieve avalanche. In those two iterations,
1840 * we have six layers of mixing, which is enough to spread one bit's
1841 * influence out to 2^6 = 64 state bits.
1843 * Rotate constants are scored by considering either 64 one-bit input
1844 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1845 * probability of that delta causing a change to each of the 128 output
1846 * bits, using a sample of random initial states.
1848 * The Shannon entropy of the computed probabilities is then summed
1849 * to produce a score. Ideally, any input change has a 50% chance of
1850 * toggling any given output bit.
1852 * Mixing scores (in bits) for (12,45):
1853 * Input delta: 1-bit 2-bit
1854 * 1 round: 713.3 42542.6
1855 * 2 rounds: 2753.7 140389.8
1856 * 3 rounds: 5954.1 233458.2
1857 * 4 rounds: 7862.6 256672.2
1858 * Perfect: 8192 258048
1859 * (64*128) (64*63/2 * 128)
1861 #define HASH_MIX(x, y, a) \
1863 y ^= x, x = rol64(x,12),\
1864 x += y, y = rol64(y,45),\
1868 * Fold two longs into one 32-bit hash value. This must be fast, but
1869 * latency isn't quite as critical, as there is a fair bit of additional
1870 * work done before the hash value is used.
1872 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1874 y ^= x * GOLDEN_RATIO_64;
1875 y *= GOLDEN_RATIO_64;
1879 #else /* 32-bit case */
1882 * Mixing scores (in bits) for (7,20):
1883 * Input delta: 1-bit 2-bit
1884 * 1 round: 330.3 9201.6
1885 * 2 rounds: 1246.4 25475.4
1886 * 3 rounds: 1907.1 31295.1
1887 * 4 rounds: 2042.3 31718.6
1888 * Perfect: 2048 31744
1889 * (32*64) (32*31/2 * 64)
1891 #define HASH_MIX(x, y, a) \
1893 y ^= x, x = rol32(x, 7),\
1894 x += y, y = rol32(y,20),\
1897 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1899 /* Use arch-optimized multiply if one exists */
1900 return __hash_32(y ^ __hash_32(x));
1906 * Return the hash of a string of known length. This is carfully
1907 * designed to match hash_name(), which is the more critical function.
1908 * In particular, we must end by hashing a final word containing 0..7
1909 * payload bytes, to match the way that hash_name() iterates until it
1910 * finds the delimiter after the name.
1912 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1914 unsigned long a, x = 0, y = (unsigned long)salt;
1919 a = load_unaligned_zeropad(name);
1920 if (len < sizeof(unsigned long))
1923 name += sizeof(unsigned long);
1924 len -= sizeof(unsigned long);
1926 x ^= a & bytemask_from_count(len);
1928 return fold_hash(x, y);
1930 EXPORT_SYMBOL(full_name_hash);
1932 /* Return the "hash_len" (hash and length) of a null-terminated string */
1933 u64 hashlen_string(const void *salt, const char *name)
1935 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1936 unsigned long adata, mask, len;
1937 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1944 len += sizeof(unsigned long);
1946 a = load_unaligned_zeropad(name+len);
1947 } while (!has_zero(a, &adata, &constants));
1949 adata = prep_zero_mask(a, adata, &constants);
1950 mask = create_zero_mask(adata);
1951 x ^= a & zero_bytemask(mask);
1953 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1955 EXPORT_SYMBOL(hashlen_string);
1958 * Calculate the length and hash of the path component, and
1959 * return the "hash_len" as the result.
1961 static inline u64 hash_name(const void *salt, const char *name)
1963 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1964 unsigned long adata, bdata, mask, len;
1965 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1972 len += sizeof(unsigned long);
1974 a = load_unaligned_zeropad(name+len);
1975 b = a ^ REPEAT_BYTE('/');
1976 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1978 adata = prep_zero_mask(a, adata, &constants);
1979 bdata = prep_zero_mask(b, bdata, &constants);
1980 mask = create_zero_mask(adata | bdata);
1981 x ^= a & zero_bytemask(mask);
1983 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1986 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1988 /* Return the hash of a string of known length */
1989 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1991 unsigned long hash = init_name_hash(salt);
1993 hash = partial_name_hash((unsigned char)*name++, hash);
1994 return end_name_hash(hash);
1996 EXPORT_SYMBOL(full_name_hash);
1998 /* Return the "hash_len" (hash and length) of a null-terminated string */
1999 u64 hashlen_string(const void *salt, const char *name)
2001 unsigned long hash = init_name_hash(salt);
2002 unsigned long len = 0, c;
2004 c = (unsigned char)*name;
2007 hash = partial_name_hash(c, hash);
2008 c = (unsigned char)name[len];
2010 return hashlen_create(end_name_hash(hash), len);
2012 EXPORT_SYMBOL(hashlen_string);
2015 * We know there's a real path component here of at least
2018 static inline u64 hash_name(const void *salt, const char *name)
2020 unsigned long hash = init_name_hash(salt);
2021 unsigned long len = 0, c;
2023 c = (unsigned char)*name;
2026 hash = partial_name_hash(c, hash);
2027 c = (unsigned char)name[len];
2028 } while (c && c != '/');
2029 return hashlen_create(end_name_hash(hash), len);
2036 * This is the basic name resolution function, turning a pathname into
2037 * the final dentry. We expect 'base' to be positive and a directory.
2039 * Returns 0 and nd will have valid dentry and mnt on success.
2040 * Returns error and drops reference to input namei data on failure.
2042 static int link_path_walk(const char *name, struct nameidata *nd)
2051 /* At this point we know we have a real path component. */
2056 err = may_lookup(nd);
2060 hash_len = hash_name(nd->path.dentry, name);
2063 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2065 if (name[1] == '.') {
2067 nd->flags |= LOOKUP_JUMPED;
2073 if (likely(type == LAST_NORM)) {
2074 struct dentry *parent = nd->path.dentry;
2075 nd->flags &= ~LOOKUP_JUMPED;
2076 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2077 struct qstr this = { { .hash_len = hash_len }, .name = name };
2078 err = parent->d_op->d_hash(parent, &this);
2081 hash_len = this.hash_len;
2086 nd->last.hash_len = hash_len;
2087 nd->last.name = name;
2088 nd->last_type = type;
2090 name += hashlen_len(hash_len);
2094 * If it wasn't NUL, we know it was '/'. Skip that
2095 * slash, and continue until no more slashes.
2099 } while (unlikely(*name == '/'));
2100 if (unlikely(!*name)) {
2102 /* pathname body, done */
2105 name = nd->stack[nd->depth - 1].name;
2106 /* trailing symlink, done */
2109 /* last component of nested symlink */
2110 err = walk_component(nd, WALK_FOLLOW);
2112 /* not the last component */
2113 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2119 const char *s = get_link(nd);
2128 nd->stack[nd->depth - 1].name = name;
2133 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2134 if (nd->flags & LOOKUP_RCU) {
2135 if (unlazy_walk(nd))
2143 static const char *path_init(struct nameidata *nd, unsigned flags)
2145 const char *s = nd->name->name;
2148 flags &= ~LOOKUP_RCU;
2150 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2151 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2153 if (flags & LOOKUP_ROOT) {
2154 struct dentry *root = nd->root.dentry;
2155 struct inode *inode = root->d_inode;
2156 if (*s && unlikely(!d_can_lookup(root)))
2157 return ERR_PTR(-ENOTDIR);
2158 nd->path = nd->root;
2160 if (flags & LOOKUP_RCU) {
2162 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2163 nd->root_seq = nd->seq;
2164 nd->m_seq = read_seqbegin(&mount_lock);
2166 path_get(&nd->path);
2171 nd->root.mnt = NULL;
2172 nd->path.mnt = NULL;
2173 nd->path.dentry = NULL;
2175 nd->m_seq = read_seqbegin(&mount_lock);
2177 if (flags & LOOKUP_RCU)
2180 if (likely(!nd_jump_root(nd)))
2182 nd->root.mnt = NULL;
2184 return ERR_PTR(-ECHILD);
2185 } else if (nd->dfd == AT_FDCWD) {
2186 if (flags & LOOKUP_RCU) {
2187 struct fs_struct *fs = current->fs;
2193 seq = read_seqcount_begin(&fs->seq);
2195 nd->inode = nd->path.dentry->d_inode;
2196 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2197 } while (read_seqcount_retry(&fs->seq, seq));
2199 get_fs_pwd(current->fs, &nd->path);
2200 nd->inode = nd->path.dentry->d_inode;
2204 /* Caller must check execute permissions on the starting path component */
2205 struct fd f = fdget_raw(nd->dfd);
2206 struct dentry *dentry;
2209 return ERR_PTR(-EBADF);
2211 dentry = f.file->f_path.dentry;
2214 if (!d_can_lookup(dentry)) {
2216 return ERR_PTR(-ENOTDIR);
2220 nd->path = f.file->f_path;
2221 if (flags & LOOKUP_RCU) {
2223 nd->inode = nd->path.dentry->d_inode;
2224 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2226 path_get(&nd->path);
2227 nd->inode = nd->path.dentry->d_inode;
2234 static const char *trailing_symlink(struct nameidata *nd)
2237 int error = may_follow_link(nd);
2238 if (unlikely(error))
2239 return ERR_PTR(error);
2240 nd->flags |= LOOKUP_PARENT;
2241 nd->stack[0].name = NULL;
2246 static inline int lookup_last(struct nameidata *nd)
2248 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2249 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2251 nd->flags &= ~LOOKUP_PARENT;
2252 return walk_component(nd, 0);
2255 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2256 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2258 const char *s = path_init(nd, flags);
2263 while (!(err = link_path_walk(s, nd))
2264 && ((err = lookup_last(nd)) > 0)) {
2265 s = trailing_symlink(nd);
2272 err = complete_walk(nd);
2274 if (!err && nd->flags & LOOKUP_DIRECTORY)
2275 if (!d_can_lookup(nd->path.dentry))
2279 nd->path.mnt = NULL;
2280 nd->path.dentry = NULL;
2286 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2287 struct path *path, struct path *root)
2290 struct nameidata nd;
2292 return PTR_ERR(name);
2293 if (unlikely(root)) {
2295 flags |= LOOKUP_ROOT;
2297 set_nameidata(&nd, dfd, name);
2298 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2299 if (unlikely(retval == -ECHILD))
2300 retval = path_lookupat(&nd, flags, path);
2301 if (unlikely(retval == -ESTALE))
2302 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2304 if (likely(!retval))
2305 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2306 restore_nameidata();
2311 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2312 static int path_parentat(struct nameidata *nd, unsigned flags,
2313 struct path *parent)
2315 const char *s = path_init(nd, flags);
2319 err = link_path_walk(s, nd);
2321 err = complete_walk(nd);
2324 nd->path.mnt = NULL;
2325 nd->path.dentry = NULL;
2331 static struct filename *filename_parentat(int dfd, struct filename *name,
2332 unsigned int flags, struct path *parent,
2333 struct qstr *last, int *type)
2336 struct nameidata nd;
2340 set_nameidata(&nd, dfd, name);
2341 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2342 if (unlikely(retval == -ECHILD))
2343 retval = path_parentat(&nd, flags, parent);
2344 if (unlikely(retval == -ESTALE))
2345 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2346 if (likely(!retval)) {
2348 *type = nd.last_type;
2349 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2352 name = ERR_PTR(retval);
2354 restore_nameidata();
2358 /* does lookup, returns the object with parent locked */
2359 struct dentry *kern_path_locked(const char *name, struct path *path)
2361 struct filename *filename;
2366 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2368 if (IS_ERR(filename))
2369 return ERR_CAST(filename);
2370 if (unlikely(type != LAST_NORM)) {
2373 return ERR_PTR(-EINVAL);
2375 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2376 d = __lookup_hash(&last, path->dentry, 0);
2378 inode_unlock(path->dentry->d_inode);
2385 int kern_path(const char *name, unsigned int flags, struct path *path)
2387 return filename_lookup(AT_FDCWD, getname_kernel(name),
2390 EXPORT_SYMBOL(kern_path);
2393 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2394 * @dentry: pointer to dentry of the base directory
2395 * @mnt: pointer to vfs mount of the base directory
2396 * @name: pointer to file name
2397 * @flags: lookup flags
2398 * @path: pointer to struct path to fill
2400 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2401 const char *name, unsigned int flags,
2404 struct path root = {.mnt = mnt, .dentry = dentry};
2405 /* the first argument of filename_lookup() is ignored with root */
2406 return filename_lookup(AT_FDCWD, getname_kernel(name),
2407 flags , path, &root);
2409 EXPORT_SYMBOL(vfs_path_lookup);
2412 * lookup_one_len - filesystem helper to lookup single pathname component
2413 * @name: pathname component to lookup
2414 * @base: base directory to lookup from
2415 * @len: maximum length @len should be interpreted to
2417 * Note that this routine is purely a helper for filesystem usage and should
2418 * not be called by generic code.
2420 * The caller must hold base->i_mutex.
2422 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2428 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2432 this.hash = full_name_hash(base, name, len);
2434 return ERR_PTR(-EACCES);
2436 if (unlikely(name[0] == '.')) {
2437 if (len < 2 || (len == 2 && name[1] == '.'))
2438 return ERR_PTR(-EACCES);
2442 c = *(const unsigned char *)name++;
2443 if (c == '/' || c == '\0')
2444 return ERR_PTR(-EACCES);
2447 * See if the low-level filesystem might want
2448 * to use its own hash..
2450 if (base->d_flags & DCACHE_OP_HASH) {
2451 int err = base->d_op->d_hash(base, &this);
2453 return ERR_PTR(err);
2456 err = inode_permission(base->d_inode, MAY_EXEC);
2458 return ERR_PTR(err);
2460 return __lookup_hash(&this, base, 0);
2462 EXPORT_SYMBOL(lookup_one_len);
2465 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2466 * @name: pathname component to lookup
2467 * @base: base directory to lookup from
2468 * @len: maximum length @len should be interpreted to
2470 * Note that this routine is purely a helper for filesystem usage and should
2471 * not be called by generic code.
2473 * Unlike lookup_one_len, it should be called without the parent
2474 * i_mutex held, and will take the i_mutex itself if necessary.
2476 struct dentry *lookup_one_len_unlocked(const char *name,
2477 struct dentry *base, int len)
2486 this.hash = full_name_hash(base, name, len);
2488 return ERR_PTR(-EACCES);
2490 if (unlikely(name[0] == '.')) {
2491 if (len < 2 || (len == 2 && name[1] == '.'))
2492 return ERR_PTR(-EACCES);
2496 c = *(const unsigned char *)name++;
2497 if (c == '/' || c == '\0')
2498 return ERR_PTR(-EACCES);
2501 * See if the low-level filesystem might want
2502 * to use its own hash..
2504 if (base->d_flags & DCACHE_OP_HASH) {
2505 int err = base->d_op->d_hash(base, &this);
2507 return ERR_PTR(err);
2510 err = inode_permission(base->d_inode, MAY_EXEC);
2512 return ERR_PTR(err);
2514 ret = lookup_dcache(&this, base, 0);
2516 ret = lookup_slow(&this, base, 0);
2519 EXPORT_SYMBOL(lookup_one_len_unlocked);
2521 #ifdef CONFIG_UNIX98_PTYS
2522 int path_pts(struct path *path)
2524 /* Find something mounted on "pts" in the same directory as
2527 struct dentry *child, *parent;
2531 ret = path_parent_directory(path);
2535 parent = path->dentry;
2538 child = d_hash_and_lookup(parent, &this);
2542 path->dentry = child;
2549 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2550 struct path *path, int *empty)
2552 return filename_lookup(dfd, getname_flags(name, flags, empty),
2555 EXPORT_SYMBOL(user_path_at_empty);
2558 * mountpoint_last - look up last component for umount
2559 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2561 * This is a special lookup_last function just for umount. In this case, we
2562 * need to resolve the path without doing any revalidation.
2564 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2565 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2566 * in almost all cases, this lookup will be served out of the dcache. The only
2567 * cases where it won't are if nd->last refers to a symlink or the path is
2568 * bogus and it doesn't exist.
2571 * -error: if there was an error during lookup. This includes -ENOENT if the
2572 * lookup found a negative dentry.
2574 * 0: if we successfully resolved nd->last and found it to not to be a
2575 * symlink that needs to be followed.
2577 * 1: if we successfully resolved nd->last and found it to be a symlink
2578 * that needs to be followed.
2581 mountpoint_last(struct nameidata *nd)
2584 struct dentry *dir = nd->path.dentry;
2587 /* If we're in rcuwalk, drop out of it to handle last component */
2588 if (nd->flags & LOOKUP_RCU) {
2589 if (unlazy_walk(nd))
2593 nd->flags &= ~LOOKUP_PARENT;
2595 if (unlikely(nd->last_type != LAST_NORM)) {
2596 error = handle_dots(nd, nd->last_type);
2599 path.dentry = dget(nd->path.dentry);
2601 path.dentry = d_lookup(dir, &nd->last);
2604 * No cached dentry. Mounted dentries are pinned in the
2605 * cache, so that means that this dentry is probably
2606 * a symlink or the path doesn't actually point
2607 * to a mounted dentry.
2609 path.dentry = lookup_slow(&nd->last, dir,
2610 nd->flags | LOOKUP_NO_REVAL);
2611 if (IS_ERR(path.dentry))
2612 return PTR_ERR(path.dentry);
2615 if (d_is_negative(path.dentry)) {
2619 path.mnt = nd->path.mnt;
2620 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2624 * path_mountpoint - look up a path to be umounted
2625 * @nd: lookup context
2626 * @flags: lookup flags
2627 * @path: pointer to container for result
2629 * Look up the given name, but don't attempt to revalidate the last component.
2630 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2633 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2635 const char *s = path_init(nd, flags);
2639 while (!(err = link_path_walk(s, nd)) &&
2640 (err = mountpoint_last(nd)) > 0) {
2641 s = trailing_symlink(nd);
2649 nd->path.mnt = NULL;
2650 nd->path.dentry = NULL;
2658 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2661 struct nameidata nd;
2664 return PTR_ERR(name);
2665 set_nameidata(&nd, dfd, name);
2666 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2667 if (unlikely(error == -ECHILD))
2668 error = path_mountpoint(&nd, flags, path);
2669 if (unlikely(error == -ESTALE))
2670 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2672 audit_inode(name, path->dentry, 0);
2673 restore_nameidata();
2679 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2680 * @dfd: directory file descriptor
2681 * @name: pathname from userland
2682 * @flags: lookup flags
2683 * @path: pointer to container to hold result
2685 * A umount is a special case for path walking. We're not actually interested
2686 * in the inode in this situation, and ESTALE errors can be a problem. We
2687 * simply want track down the dentry and vfsmount attached at the mountpoint
2688 * and avoid revalidating the last component.
2690 * Returns 0 and populates "path" on success.
2693 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2696 return filename_mountpoint(dfd, getname(name), path, flags);
2700 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2703 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2705 EXPORT_SYMBOL(kern_path_mountpoint);
2707 int __check_sticky(struct inode *dir, struct inode *inode)
2709 kuid_t fsuid = current_fsuid();
2711 if (uid_eq(inode->i_uid, fsuid))
2713 if (uid_eq(dir->i_uid, fsuid))
2715 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2717 EXPORT_SYMBOL(__check_sticky);
2720 * Check whether we can remove a link victim from directory dir, check
2721 * whether the type of victim is right.
2722 * 1. We can't do it if dir is read-only (done in permission())
2723 * 2. We should have write and exec permissions on dir
2724 * 3. We can't remove anything from append-only dir
2725 * 4. We can't do anything with immutable dir (done in permission())
2726 * 5. If the sticky bit on dir is set we should either
2727 * a. be owner of dir, or
2728 * b. be owner of victim, or
2729 * c. have CAP_FOWNER capability
2730 * 6. If the victim is append-only or immutable we can't do antyhing with
2731 * links pointing to it.
2732 * 7. If the victim has an unknown uid or gid we can't change the inode.
2733 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2734 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2735 * 10. We can't remove a root or mountpoint.
2736 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2737 * nfs_async_unlink().
2739 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2741 struct inode *inode = d_backing_inode(victim);
2744 if (d_is_negative(victim))
2748 BUG_ON(victim->d_parent->d_inode != dir);
2749 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2751 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2757 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2758 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2761 if (!d_is_dir(victim))
2763 if (IS_ROOT(victim))
2765 } else if (d_is_dir(victim))
2767 if (IS_DEADDIR(dir))
2769 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2774 /* Check whether we can create an object with dentry child in directory
2776 * 1. We can't do it if child already exists (open has special treatment for
2777 * this case, but since we are inlined it's OK)
2778 * 2. We can't do it if dir is read-only (done in permission())
2779 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2780 * 4. We should have write and exec permissions on dir
2781 * 5. We can't do it if dir is immutable (done in permission())
2783 static inline int may_create(struct inode *dir, struct dentry *child)
2785 struct user_namespace *s_user_ns;
2786 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2789 if (IS_DEADDIR(dir))
2791 s_user_ns = dir->i_sb->s_user_ns;
2792 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2793 !kgid_has_mapping(s_user_ns, current_fsgid()))
2795 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2799 * p1 and p2 should be directories on the same fs.
2801 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2806 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2810 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2812 p = d_ancestor(p2, p1);
2814 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2815 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2819 p = d_ancestor(p1, p2);
2821 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2822 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2826 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2827 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2830 EXPORT_SYMBOL(lock_rename);
2832 void unlock_rename(struct dentry *p1, struct dentry *p2)
2834 inode_unlock(p1->d_inode);
2836 inode_unlock(p2->d_inode);
2837 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2840 EXPORT_SYMBOL(unlock_rename);
2842 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2845 int error = may_create(dir, dentry);
2849 if (!dir->i_op->create)
2850 return -EACCES; /* shouldn't it be ENOSYS? */
2853 error = security_inode_create(dir, dentry, mode);
2856 error = dir->i_op->create(dir, dentry, mode, want_excl);
2858 fsnotify_create(dir, dentry);
2861 EXPORT_SYMBOL(vfs_create);
2863 bool may_open_dev(const struct path *path)
2865 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2866 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2869 static int may_open(const struct path *path, int acc_mode, int flag)
2871 struct dentry *dentry = path->dentry;
2872 struct inode *inode = dentry->d_inode;
2878 switch (inode->i_mode & S_IFMT) {
2882 if (acc_mode & MAY_WRITE)
2887 if (!may_open_dev(path))
2896 error = inode_permission(inode, MAY_OPEN | acc_mode);
2901 * An append-only file must be opened in append mode for writing.
2903 if (IS_APPEND(inode)) {
2904 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2910 /* O_NOATIME can only be set by the owner or superuser */
2911 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2917 static int handle_truncate(struct file *filp)
2919 const struct path *path = &filp->f_path;
2920 struct inode *inode = path->dentry->d_inode;
2921 int error = get_write_access(inode);
2925 * Refuse to truncate files with mandatory locks held on them.
2927 error = locks_verify_locked(filp);
2929 error = security_path_truncate(path);
2931 error = do_truncate(path->dentry, 0,
2932 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2935 put_write_access(inode);
2939 static inline int open_to_namei_flags(int flag)
2941 if ((flag & O_ACCMODE) == 3)
2946 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2948 struct user_namespace *s_user_ns;
2949 int error = security_path_mknod(dir, dentry, mode, 0);
2953 s_user_ns = dir->dentry->d_sb->s_user_ns;
2954 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2955 !kgid_has_mapping(s_user_ns, current_fsgid()))
2958 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2962 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2966 * Attempt to atomically look up, create and open a file from a negative
2969 * Returns 0 if successful. The file will have been created and attached to
2970 * @file by the filesystem calling finish_open().
2972 * Returns 1 if the file was looked up only or didn't need creating. The
2973 * caller will need to perform the open themselves. @path will have been
2974 * updated to point to the new dentry. This may be negative.
2976 * Returns an error code otherwise.
2978 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2979 struct path *path, struct file *file,
2980 const struct open_flags *op,
2981 int open_flag, umode_t mode,
2984 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2985 struct inode *dir = nd->path.dentry->d_inode;
2988 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
2989 open_flag &= ~O_TRUNC;
2991 if (nd->flags & LOOKUP_DIRECTORY)
2992 open_flag |= O_DIRECTORY;
2994 file->f_path.dentry = DENTRY_NOT_SET;
2995 file->f_path.mnt = nd->path.mnt;
2996 error = dir->i_op->atomic_open(dir, dentry, file,
2997 open_to_namei_flags(open_flag),
2999 d_lookup_done(dentry);
3002 * We didn't have the inode before the open, so check open
3005 int acc_mode = op->acc_mode;
3006 if (*opened & FILE_CREATED) {
3007 WARN_ON(!(open_flag & O_CREAT));
3008 fsnotify_create(dir, dentry);
3011 error = may_open(&file->f_path, acc_mode, open_flag);
3012 if (WARN_ON(error > 0))
3014 } else if (error > 0) {
3015 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3018 if (file->f_path.dentry) {
3020 dentry = file->f_path.dentry;
3022 if (*opened & FILE_CREATED)
3023 fsnotify_create(dir, dentry);
3024 if (unlikely(d_is_negative(dentry))) {
3027 path->dentry = dentry;
3028 path->mnt = nd->path.mnt;
3038 * Look up and maybe create and open the last component.
3040 * Must be called with i_mutex held on parent.
3042 * Returns 0 if the file was successfully atomically created (if necessary) and
3043 * opened. In this case the file will be returned attached to @file.
3045 * Returns 1 if the file was not completely opened at this time, though lookups
3046 * and creations will have been performed and the dentry returned in @path will
3047 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3048 * specified then a negative dentry may be returned.
3050 * An error code is returned otherwise.
3052 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3053 * cleared otherwise prior to returning.
3055 static int lookup_open(struct nameidata *nd, struct path *path,
3057 const struct open_flags *op,
3058 bool got_write, int *opened)
3060 struct dentry *dir = nd->path.dentry;
3061 struct inode *dir_inode = dir->d_inode;
3062 int open_flag = op->open_flag;
3063 struct dentry *dentry;
3064 int error, create_error = 0;
3065 umode_t mode = op->mode;
3066 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3068 if (unlikely(IS_DEADDIR(dir_inode)))
3071 *opened &= ~FILE_CREATED;
3072 dentry = d_lookup(dir, &nd->last);
3075 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3077 return PTR_ERR(dentry);
3079 if (d_in_lookup(dentry))
3082 error = d_revalidate(dentry, nd->flags);
3083 if (likely(error > 0))
3087 d_invalidate(dentry);
3091 if (dentry->d_inode) {
3092 /* Cached positive dentry: will open in f_op->open */
3097 * Checking write permission is tricky, bacuse we don't know if we are
3098 * going to actually need it: O_CREAT opens should work as long as the
3099 * file exists. But checking existence breaks atomicity. The trick is
3100 * to check access and if not granted clear O_CREAT from the flags.
3102 * Another problem is returing the "right" error value (e.g. for an
3103 * O_EXCL open we want to return EEXIST not EROFS).
3105 if (open_flag & O_CREAT) {
3106 if (!IS_POSIXACL(dir->d_inode))
3107 mode &= ~current_umask();
3108 if (unlikely(!got_write)) {
3109 create_error = -EROFS;
3110 open_flag &= ~O_CREAT;
3111 if (open_flag & (O_EXCL | O_TRUNC))
3113 /* No side effects, safe to clear O_CREAT */
3115 create_error = may_o_create(&nd->path, dentry, mode);
3117 open_flag &= ~O_CREAT;
3118 if (open_flag & O_EXCL)
3122 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3123 unlikely(!got_write)) {
3125 * No O_CREATE -> atomicity not a requirement -> fall
3126 * back to lookup + open
3131 if (dir_inode->i_op->atomic_open) {
3132 error = atomic_open(nd, dentry, path, file, op, open_flag,
3134 if (unlikely(error == -ENOENT) && create_error)
3135 error = create_error;
3140 if (d_in_lookup(dentry)) {
3141 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3143 d_lookup_done(dentry);
3144 if (unlikely(res)) {
3146 error = PTR_ERR(res);
3154 /* Negative dentry, just create the file */
3155 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3156 *opened |= FILE_CREATED;
3157 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3158 if (!dir_inode->i_op->create) {
3162 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3163 open_flag & O_EXCL);
3166 fsnotify_create(dir_inode, dentry);
3168 if (unlikely(create_error) && !dentry->d_inode) {
3169 error = create_error;
3173 path->dentry = dentry;
3174 path->mnt = nd->path.mnt;
3183 * Handle the last step of open()
3185 static int do_last(struct nameidata *nd,
3186 struct file *file, const struct open_flags *op,
3189 struct dentry *dir = nd->path.dentry;
3190 int open_flag = op->open_flag;
3191 bool will_truncate = (open_flag & O_TRUNC) != 0;
3192 bool got_write = false;
3193 int acc_mode = op->acc_mode;
3195 struct inode *inode;
3199 nd->flags &= ~LOOKUP_PARENT;
3200 nd->flags |= op->intent;
3202 if (nd->last_type != LAST_NORM) {
3203 error = handle_dots(nd, nd->last_type);
3204 if (unlikely(error))
3209 if (!(open_flag & O_CREAT)) {
3210 if (nd->last.name[nd->last.len])
3211 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3212 /* we _can_ be in RCU mode here */
3213 error = lookup_fast(nd, &path, &inode, &seq);
3214 if (likely(error > 0))
3220 BUG_ON(nd->inode != dir->d_inode);
3221 BUG_ON(nd->flags & LOOKUP_RCU);
3223 /* create side of things */
3225 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3226 * has been cleared when we got to the last component we are
3229 error = complete_walk(nd);
3233 audit_inode(nd->name, dir, LOOKUP_PARENT);
3234 /* trailing slashes? */
3235 if (unlikely(nd->last.name[nd->last.len]))
3239 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3240 error = mnt_want_write(nd->path.mnt);
3244 * do _not_ fail yet - we might not need that or fail with
3245 * a different error; let lookup_open() decide; we'll be
3246 * dropping this one anyway.
3249 if (open_flag & O_CREAT)
3250 inode_lock(dir->d_inode);
3252 inode_lock_shared(dir->d_inode);
3253 error = lookup_open(nd, &path, file, op, got_write, opened);
3254 if (open_flag & O_CREAT)
3255 inode_unlock(dir->d_inode);
3257 inode_unlock_shared(dir->d_inode);
3263 if ((*opened & FILE_CREATED) ||
3264 !S_ISREG(file_inode(file)->i_mode))
3265 will_truncate = false;
3267 audit_inode(nd->name, file->f_path.dentry, 0);
3271 if (*opened & FILE_CREATED) {
3272 /* Don't check for write permission, don't truncate */
3273 open_flag &= ~O_TRUNC;
3274 will_truncate = false;
3276 path_to_nameidata(&path, nd);
3277 goto finish_open_created;
3281 * If atomic_open() acquired write access it is dropped now due to
3282 * possible mount and symlink following (this might be optimized away if
3286 mnt_drop_write(nd->path.mnt);
3290 error = follow_managed(&path, nd);
3291 if (unlikely(error < 0))
3294 if (unlikely(d_is_negative(path.dentry))) {
3295 path_to_nameidata(&path, nd);
3300 * create/update audit record if it already exists.
3302 audit_inode(nd->name, path.dentry, 0);
3304 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3305 path_to_nameidata(&path, nd);
3309 seq = 0; /* out of RCU mode, so the value doesn't matter */
3310 inode = d_backing_inode(path.dentry);
3312 error = step_into(nd, &path, 0, inode, seq);
3313 if (unlikely(error))
3316 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3317 error = complete_walk(nd);
3320 audit_inode(nd->name, nd->path.dentry, 0);
3322 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3325 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3327 if (!d_is_reg(nd->path.dentry))
3328 will_truncate = false;
3330 if (will_truncate) {
3331 error = mnt_want_write(nd->path.mnt);
3336 finish_open_created:
3337 error = may_open(&nd->path, acc_mode, open_flag);
3340 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3341 error = vfs_open(&nd->path, file, current_cred());
3344 *opened |= FILE_OPENED;
3346 error = open_check_o_direct(file);
3348 error = ima_file_check(file, op->acc_mode, *opened);
3349 if (!error && will_truncate)
3350 error = handle_truncate(file);
3352 if (unlikely(error) && (*opened & FILE_OPENED))
3354 if (unlikely(error > 0)) {
3359 mnt_drop_write(nd->path.mnt);
3363 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3365 static const struct qstr name = QSTR_INIT("/", 1);
3366 struct dentry *child = NULL;
3367 struct inode *dir = dentry->d_inode;
3368 struct inode *inode;
3371 /* we want directory to be writable */
3372 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3375 error = -EOPNOTSUPP;
3376 if (!dir->i_op->tmpfile)
3379 child = d_alloc(dentry, &name);
3380 if (unlikely(!child))
3382 error = dir->i_op->tmpfile(dir, child, mode);
3386 inode = child->d_inode;
3387 if (unlikely(!inode))
3389 if (!(open_flag & O_EXCL)) {
3390 spin_lock(&inode->i_lock);
3391 inode->i_state |= I_LINKABLE;
3392 spin_unlock(&inode->i_lock);
3398 return ERR_PTR(error);
3400 EXPORT_SYMBOL(vfs_tmpfile);
3402 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3403 const struct open_flags *op,
3404 struct file *file, int *opened)
3406 struct dentry *child;
3408 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3409 if (unlikely(error))
3411 error = mnt_want_write(path.mnt);
3412 if (unlikely(error))
3414 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3415 error = PTR_ERR(child);
3416 if (unlikely(IS_ERR(child)))
3419 path.dentry = child;
3420 audit_inode(nd->name, child, 0);
3421 /* Don't check for other permissions, the inode was just created */
3422 error = may_open(&path, 0, op->open_flag);
3425 file->f_path.mnt = path.mnt;
3426 error = finish_open(file, child, NULL, opened);
3429 error = open_check_o_direct(file);
3433 mnt_drop_write(path.mnt);
3439 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3442 int error = path_lookupat(nd, flags, &path);
3444 audit_inode(nd->name, path.dentry, 0);
3445 error = vfs_open(&path, file, current_cred());
3451 static struct file *path_openat(struct nameidata *nd,
3452 const struct open_flags *op, unsigned flags)
3459 file = get_empty_filp();
3463 file->f_flags = op->open_flag;
3465 if (unlikely(file->f_flags & __O_TMPFILE)) {
3466 error = do_tmpfile(nd, flags, op, file, &opened);
3470 if (unlikely(file->f_flags & O_PATH)) {
3471 error = do_o_path(nd, flags, file);
3473 opened |= FILE_OPENED;
3477 s = path_init(nd, flags);
3482 while (!(error = link_path_walk(s, nd)) &&
3483 (error = do_last(nd, file, op, &opened)) > 0) {
3484 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3485 s = trailing_symlink(nd);
3493 if (!(opened & FILE_OPENED)) {
3497 if (unlikely(error)) {
3498 if (error == -EOPENSTALE) {
3499 if (flags & LOOKUP_RCU)
3504 file = ERR_PTR(error);
3509 struct file *do_filp_open(int dfd, struct filename *pathname,
3510 const struct open_flags *op)
3512 struct nameidata nd;
3513 int flags = op->lookup_flags;
3516 set_nameidata(&nd, dfd, pathname);
3517 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3518 if (unlikely(filp == ERR_PTR(-ECHILD)))
3519 filp = path_openat(&nd, op, flags);
3520 if (unlikely(filp == ERR_PTR(-ESTALE)))
3521 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3522 restore_nameidata();
3526 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3527 const char *name, const struct open_flags *op)
3529 struct nameidata nd;
3531 struct filename *filename;
3532 int flags = op->lookup_flags | LOOKUP_ROOT;
3535 nd.root.dentry = dentry;
3537 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3538 return ERR_PTR(-ELOOP);
3540 filename = getname_kernel(name);
3541 if (IS_ERR(filename))
3542 return ERR_CAST(filename);
3544 set_nameidata(&nd, -1, filename);
3545 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3546 if (unlikely(file == ERR_PTR(-ECHILD)))
3547 file = path_openat(&nd, op, flags);
3548 if (unlikely(file == ERR_PTR(-ESTALE)))
3549 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3550 restore_nameidata();
3555 static struct dentry *filename_create(int dfd, struct filename *name,
3556 struct path *path, unsigned int lookup_flags)
3558 struct dentry *dentry = ERR_PTR(-EEXIST);
3563 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3566 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3567 * other flags passed in are ignored!
3569 lookup_flags &= LOOKUP_REVAL;
3571 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3573 return ERR_CAST(name);
3576 * Yucky last component or no last component at all?
3577 * (foo/., foo/.., /////)
3579 if (unlikely(type != LAST_NORM))
3582 /* don't fail immediately if it's r/o, at least try to report other errors */
3583 err2 = mnt_want_write(path->mnt);
3585 * Do the final lookup.
3587 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3588 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3589 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3594 if (d_is_positive(dentry))
3598 * Special case - lookup gave negative, but... we had foo/bar/
3599 * From the vfs_mknod() POV we just have a negative dentry -
3600 * all is fine. Let's be bastards - you had / on the end, you've
3601 * been asking for (non-existent) directory. -ENOENT for you.
3603 if (unlikely(!is_dir && last.name[last.len])) {
3607 if (unlikely(err2)) {
3615 dentry = ERR_PTR(error);
3617 inode_unlock(path->dentry->d_inode);
3619 mnt_drop_write(path->mnt);
3626 struct dentry *kern_path_create(int dfd, const char *pathname,
3627 struct path *path, unsigned int lookup_flags)
3629 return filename_create(dfd, getname_kernel(pathname),
3630 path, lookup_flags);
3632 EXPORT_SYMBOL(kern_path_create);
3634 void done_path_create(struct path *path, struct dentry *dentry)
3637 inode_unlock(path->dentry->d_inode);
3638 mnt_drop_write(path->mnt);
3641 EXPORT_SYMBOL(done_path_create);
3643 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3644 struct path *path, unsigned int lookup_flags)
3646 return filename_create(dfd, getname(pathname), path, lookup_flags);
3648 EXPORT_SYMBOL(user_path_create);
3650 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3652 int error = may_create(dir, dentry);
3657 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3660 if (!dir->i_op->mknod)
3663 error = devcgroup_inode_mknod(mode, dev);
3667 error = security_inode_mknod(dir, dentry, mode, dev);
3671 error = dir->i_op->mknod(dir, dentry, mode, dev);
3673 fsnotify_create(dir, dentry);
3676 EXPORT_SYMBOL(vfs_mknod);
3678 static int may_mknod(umode_t mode)
3680 switch (mode & S_IFMT) {
3686 case 0: /* zero mode translates to S_IFREG */
3695 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3698 struct dentry *dentry;
3701 unsigned int lookup_flags = 0;
3703 error = may_mknod(mode);
3707 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3709 return PTR_ERR(dentry);
3711 if (!IS_POSIXACL(path.dentry->d_inode))
3712 mode &= ~current_umask();
3713 error = security_path_mknod(&path, dentry, mode, dev);
3716 switch (mode & S_IFMT) {
3717 case 0: case S_IFREG:
3718 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3720 ima_post_path_mknod(dentry);
3722 case S_IFCHR: case S_IFBLK:
3723 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3724 new_decode_dev(dev));
3726 case S_IFIFO: case S_IFSOCK:
3727 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3731 done_path_create(&path, dentry);
3732 if (retry_estale(error, lookup_flags)) {
3733 lookup_flags |= LOOKUP_REVAL;
3739 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3741 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3744 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3746 int error = may_create(dir, dentry);
3747 unsigned max_links = dir->i_sb->s_max_links;
3752 if (!dir->i_op->mkdir)
3755 mode &= (S_IRWXUGO|S_ISVTX);
3756 error = security_inode_mkdir(dir, dentry, mode);
3760 if (max_links && dir->i_nlink >= max_links)
3763 error = dir->i_op->mkdir(dir, dentry, mode);
3765 fsnotify_mkdir(dir, dentry);
3768 EXPORT_SYMBOL(vfs_mkdir);
3770 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3772 struct dentry *dentry;
3775 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3778 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3780 return PTR_ERR(dentry);
3782 if (!IS_POSIXACL(path.dentry->d_inode))
3783 mode &= ~current_umask();
3784 error = security_path_mkdir(&path, dentry, mode);
3786 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3787 done_path_create(&path, dentry);
3788 if (retry_estale(error, lookup_flags)) {
3789 lookup_flags |= LOOKUP_REVAL;
3795 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3797 return sys_mkdirat(AT_FDCWD, pathname, mode);
3800 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3802 int error = may_delete(dir, dentry, 1);
3807 if (!dir->i_op->rmdir)
3811 inode_lock(dentry->d_inode);
3814 if (is_local_mountpoint(dentry))
3817 error = security_inode_rmdir(dir, dentry);
3821 shrink_dcache_parent(dentry);
3822 error = dir->i_op->rmdir(dir, dentry);
3826 dentry->d_inode->i_flags |= S_DEAD;
3828 detach_mounts(dentry);
3831 inode_unlock(dentry->d_inode);
3837 EXPORT_SYMBOL(vfs_rmdir);
3839 static long do_rmdir(int dfd, const char __user *pathname)
3842 struct filename *name;
3843 struct dentry *dentry;
3847 unsigned int lookup_flags = 0;
3849 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3850 &path, &last, &type);
3852 return PTR_ERR(name);
3866 error = mnt_want_write(path.mnt);
3870 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3871 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3872 error = PTR_ERR(dentry);
3875 if (!dentry->d_inode) {
3879 error = security_path_rmdir(&path, dentry);
3882 error = vfs_rmdir(path.dentry->d_inode, dentry);
3886 inode_unlock(path.dentry->d_inode);
3887 mnt_drop_write(path.mnt);
3891 if (retry_estale(error, lookup_flags)) {
3892 lookup_flags |= LOOKUP_REVAL;
3898 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3900 return do_rmdir(AT_FDCWD, pathname);
3904 * vfs_unlink - unlink a filesystem object
3905 * @dir: parent directory
3907 * @delegated_inode: returns victim inode, if the inode is delegated.
3909 * The caller must hold dir->i_mutex.
3911 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3912 * return a reference to the inode in delegated_inode. The caller
3913 * should then break the delegation on that inode and retry. Because
3914 * breaking a delegation may take a long time, the caller should drop
3915 * dir->i_mutex before doing so.
3917 * Alternatively, a caller may pass NULL for delegated_inode. This may
3918 * be appropriate for callers that expect the underlying filesystem not
3919 * to be NFS exported.
3921 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3923 struct inode *target = dentry->d_inode;
3924 int error = may_delete(dir, dentry, 0);
3929 if (!dir->i_op->unlink)
3933 if (is_local_mountpoint(dentry))
3936 error = security_inode_unlink(dir, dentry);
3938 error = try_break_deleg(target, delegated_inode);
3941 error = dir->i_op->unlink(dir, dentry);
3944 detach_mounts(dentry);
3949 inode_unlock(target);
3951 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3952 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3953 fsnotify_link_count(target);
3959 EXPORT_SYMBOL(vfs_unlink);
3962 * Make sure that the actual truncation of the file will occur outside its
3963 * directory's i_mutex. Truncate can take a long time if there is a lot of
3964 * writeout happening, and we don't want to prevent access to the directory
3965 * while waiting on the I/O.
3967 static long do_unlinkat(int dfd, const char __user *pathname)
3970 struct filename *name;
3971 struct dentry *dentry;
3975 struct inode *inode = NULL;
3976 struct inode *delegated_inode = NULL;
3977 unsigned int lookup_flags = 0;
3979 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3980 &path, &last, &type);
3982 return PTR_ERR(name);
3985 if (type != LAST_NORM)
3988 error = mnt_want_write(path.mnt);
3992 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3993 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3994 error = PTR_ERR(dentry);
3995 if (!IS_ERR(dentry)) {
3996 /* Why not before? Because we want correct error value */
3997 if (last.name[last.len])
3999 inode = dentry->d_inode;
4000 if (d_is_negative(dentry))
4003 error = security_path_unlink(&path, dentry);
4006 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4010 inode_unlock(path.dentry->d_inode);
4012 iput(inode); /* truncate the inode here */
4014 if (delegated_inode) {
4015 error = break_deleg_wait(&delegated_inode);
4019 mnt_drop_write(path.mnt);
4023 if (retry_estale(error, lookup_flags)) {
4024 lookup_flags |= LOOKUP_REVAL;
4031 if (d_is_negative(dentry))
4033 else if (d_is_dir(dentry))
4040 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4042 if ((flag & ~AT_REMOVEDIR) != 0)
4045 if (flag & AT_REMOVEDIR)
4046 return do_rmdir(dfd, pathname);
4048 return do_unlinkat(dfd, pathname);
4051 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4053 return do_unlinkat(AT_FDCWD, pathname);
4056 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4058 int error = may_create(dir, dentry);
4063 if (!dir->i_op->symlink)
4066 error = security_inode_symlink(dir, dentry, oldname);
4070 error = dir->i_op->symlink(dir, dentry, oldname);
4072 fsnotify_create(dir, dentry);
4075 EXPORT_SYMBOL(vfs_symlink);
4077 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4078 int, newdfd, const char __user *, newname)
4081 struct filename *from;
4082 struct dentry *dentry;
4084 unsigned int lookup_flags = 0;
4086 from = getname(oldname);
4088 return PTR_ERR(from);
4090 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4091 error = PTR_ERR(dentry);
4095 error = security_path_symlink(&path, dentry, from->name);
4097 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4098 done_path_create(&path, dentry);
4099 if (retry_estale(error, lookup_flags)) {
4100 lookup_flags |= LOOKUP_REVAL;
4108 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4110 return sys_symlinkat(oldname, AT_FDCWD, newname);
4114 * vfs_link - create a new link
4115 * @old_dentry: object to be linked
4117 * @new_dentry: where to create the new link
4118 * @delegated_inode: returns inode needing a delegation break
4120 * The caller must hold dir->i_mutex
4122 * If vfs_link discovers a delegation on the to-be-linked file in need
4123 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4124 * inode in delegated_inode. The caller should then break the delegation
4125 * and retry. Because breaking a delegation may take a long time, the
4126 * caller should drop the i_mutex before doing so.
4128 * Alternatively, a caller may pass NULL for delegated_inode. This may
4129 * be appropriate for callers that expect the underlying filesystem not
4130 * to be NFS exported.
4132 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4134 struct inode *inode = old_dentry->d_inode;
4135 unsigned max_links = dir->i_sb->s_max_links;
4141 error = may_create(dir, new_dentry);
4145 if (dir->i_sb != inode->i_sb)
4149 * A link to an append-only or immutable file cannot be created.
4151 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4154 * Updating the link count will likely cause i_uid and i_gid to
4155 * be writen back improperly if their true value is unknown to
4158 if (HAS_UNMAPPED_ID(inode))
4160 if (!dir->i_op->link)
4162 if (S_ISDIR(inode->i_mode))
4165 error = security_inode_link(old_dentry, dir, new_dentry);
4170 /* Make sure we don't allow creating hardlink to an unlinked file */
4171 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4173 else if (max_links && inode->i_nlink >= max_links)
4176 error = try_break_deleg(inode, delegated_inode);
4178 error = dir->i_op->link(old_dentry, dir, new_dentry);
4181 if (!error && (inode->i_state & I_LINKABLE)) {
4182 spin_lock(&inode->i_lock);
4183 inode->i_state &= ~I_LINKABLE;
4184 spin_unlock(&inode->i_lock);
4186 inode_unlock(inode);
4188 fsnotify_link(dir, inode, new_dentry);
4191 EXPORT_SYMBOL(vfs_link);
4194 * Hardlinks are often used in delicate situations. We avoid
4195 * security-related surprises by not following symlinks on the
4198 * We don't follow them on the oldname either to be compatible
4199 * with linux 2.0, and to avoid hard-linking to directories
4200 * and other special files. --ADM
4202 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4203 int, newdfd, const char __user *, newname, int, flags)
4205 struct dentry *new_dentry;
4206 struct path old_path, new_path;
4207 struct inode *delegated_inode = NULL;
4211 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4214 * To use null names we require CAP_DAC_READ_SEARCH
4215 * This ensures that not everyone will be able to create
4216 * handlink using the passed filedescriptor.
4218 if (flags & AT_EMPTY_PATH) {
4219 if (!capable(CAP_DAC_READ_SEARCH))
4224 if (flags & AT_SYMLINK_FOLLOW)
4225 how |= LOOKUP_FOLLOW;
4227 error = user_path_at(olddfd, oldname, how, &old_path);
4231 new_dentry = user_path_create(newdfd, newname, &new_path,
4232 (how & LOOKUP_REVAL));
4233 error = PTR_ERR(new_dentry);
4234 if (IS_ERR(new_dentry))
4238 if (old_path.mnt != new_path.mnt)
4240 error = may_linkat(&old_path);
4241 if (unlikely(error))
4243 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4246 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4248 done_path_create(&new_path, new_dentry);
4249 if (delegated_inode) {
4250 error = break_deleg_wait(&delegated_inode);
4252 path_put(&old_path);
4256 if (retry_estale(error, how)) {
4257 path_put(&old_path);
4258 how |= LOOKUP_REVAL;
4262 path_put(&old_path);
4267 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4269 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4273 * vfs_rename - rename a filesystem object
4274 * @old_dir: parent of source
4275 * @old_dentry: source
4276 * @new_dir: parent of destination
4277 * @new_dentry: destination
4278 * @delegated_inode: returns an inode needing a delegation break
4279 * @flags: rename flags
4281 * The caller must hold multiple mutexes--see lock_rename()).
4283 * If vfs_rename discovers a delegation in need of breaking at either
4284 * the source or destination, it will return -EWOULDBLOCK and return a
4285 * reference to the inode in delegated_inode. The caller should then
4286 * break the delegation and retry. Because breaking a delegation may
4287 * take a long time, the caller should drop all locks before doing
4290 * Alternatively, a caller may pass NULL for delegated_inode. This may
4291 * be appropriate for callers that expect the underlying filesystem not
4292 * to be NFS exported.
4294 * The worst of all namespace operations - renaming directory. "Perverted"
4295 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4297 * a) we can get into loop creation.
4298 * b) race potential - two innocent renames can create a loop together.
4299 * That's where 4.4 screws up. Current fix: serialization on
4300 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4302 * c) we have to lock _four_ objects - parents and victim (if it exists),
4303 * and source (if it is not a directory).
4304 * And that - after we got ->i_mutex on parents (until then we don't know
4305 * whether the target exists). Solution: try to be smart with locking
4306 * order for inodes. We rely on the fact that tree topology may change
4307 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4308 * move will be locked. Thus we can rank directories by the tree
4309 * (ancestors first) and rank all non-directories after them.
4310 * That works since everybody except rename does "lock parent, lookup,
4311 * lock child" and rename is under ->s_vfs_rename_mutex.
4312 * HOWEVER, it relies on the assumption that any object with ->lookup()
4313 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4314 * we'd better make sure that there's no link(2) for them.
4315 * d) conversion from fhandle to dentry may come in the wrong moment - when
4316 * we are removing the target. Solution: we will have to grab ->i_mutex
4317 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4318 * ->i_mutex on parents, which works but leads to some truly excessive
4321 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4322 struct inode *new_dir, struct dentry *new_dentry,
4323 struct inode **delegated_inode, unsigned int flags)
4326 bool is_dir = d_is_dir(old_dentry);
4327 const unsigned char *old_name;
4328 struct inode *source = old_dentry->d_inode;
4329 struct inode *target = new_dentry->d_inode;
4330 bool new_is_dir = false;
4331 unsigned max_links = new_dir->i_sb->s_max_links;
4333 if (source == target)
4336 error = may_delete(old_dir, old_dentry, is_dir);
4341 error = may_create(new_dir, new_dentry);
4343 new_is_dir = d_is_dir(new_dentry);
4345 if (!(flags & RENAME_EXCHANGE))
4346 error = may_delete(new_dir, new_dentry, is_dir);
4348 error = may_delete(new_dir, new_dentry, new_is_dir);
4353 if (!old_dir->i_op->rename)
4357 * If we are going to change the parent - check write permissions,
4358 * we'll need to flip '..'.
4360 if (new_dir != old_dir) {
4362 error = inode_permission(source, MAY_WRITE);
4366 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4367 error = inode_permission(target, MAY_WRITE);
4373 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4378 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4380 if (!is_dir || (flags & RENAME_EXCHANGE))
4381 lock_two_nondirectories(source, target);
4386 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4389 if (max_links && new_dir != old_dir) {
4391 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4393 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4394 old_dir->i_nlink >= max_links)
4397 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4398 shrink_dcache_parent(new_dentry);
4400 error = try_break_deleg(source, delegated_inode);
4404 if (target && !new_is_dir) {
4405 error = try_break_deleg(target, delegated_inode);
4409 error = old_dir->i_op->rename(old_dir, old_dentry,
4410 new_dir, new_dentry, flags);
4414 if (!(flags & RENAME_EXCHANGE) && target) {
4416 target->i_flags |= S_DEAD;
4417 dont_mount(new_dentry);
4418 detach_mounts(new_dentry);
4420 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4421 if (!(flags & RENAME_EXCHANGE))
4422 d_move(old_dentry, new_dentry);
4424 d_exchange(old_dentry, new_dentry);
4427 if (!is_dir || (flags & RENAME_EXCHANGE))
4428 unlock_two_nondirectories(source, target);
4430 inode_unlock(target);
4433 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4434 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4435 if (flags & RENAME_EXCHANGE) {
4436 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4437 new_is_dir, NULL, new_dentry);
4440 fsnotify_oldname_free(old_name);
4444 EXPORT_SYMBOL(vfs_rename);
4446 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4447 int, newdfd, const char __user *, newname, unsigned int, flags)
4449 struct dentry *old_dentry, *new_dentry;
4450 struct dentry *trap;
4451 struct path old_path, new_path;
4452 struct qstr old_last, new_last;
4453 int old_type, new_type;
4454 struct inode *delegated_inode = NULL;
4455 struct filename *from;
4456 struct filename *to;
4457 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4458 bool should_retry = false;
4461 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4464 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4465 (flags & RENAME_EXCHANGE))
4468 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4471 if (flags & RENAME_EXCHANGE)
4475 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4476 &old_path, &old_last, &old_type);
4478 error = PTR_ERR(from);
4482 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4483 &new_path, &new_last, &new_type);
4485 error = PTR_ERR(to);
4490 if (old_path.mnt != new_path.mnt)
4494 if (old_type != LAST_NORM)
4497 if (flags & RENAME_NOREPLACE)
4499 if (new_type != LAST_NORM)
4502 error = mnt_want_write(old_path.mnt);
4507 trap = lock_rename(new_path.dentry, old_path.dentry);
4509 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4510 error = PTR_ERR(old_dentry);
4511 if (IS_ERR(old_dentry))
4513 /* source must exist */
4515 if (d_is_negative(old_dentry))
4517 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4518 error = PTR_ERR(new_dentry);
4519 if (IS_ERR(new_dentry))
4522 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4524 if (flags & RENAME_EXCHANGE) {
4526 if (d_is_negative(new_dentry))
4529 if (!d_is_dir(new_dentry)) {
4531 if (new_last.name[new_last.len])
4535 /* unless the source is a directory trailing slashes give -ENOTDIR */
4536 if (!d_is_dir(old_dentry)) {
4538 if (old_last.name[old_last.len])
4540 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4543 /* source should not be ancestor of target */
4545 if (old_dentry == trap)
4547 /* target should not be an ancestor of source */
4548 if (!(flags & RENAME_EXCHANGE))
4550 if (new_dentry == trap)
4553 error = security_path_rename(&old_path, old_dentry,
4554 &new_path, new_dentry, flags);
4557 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4558 new_path.dentry->d_inode, new_dentry,
4559 &delegated_inode, flags);
4565 unlock_rename(new_path.dentry, old_path.dentry);
4566 if (delegated_inode) {
4567 error = break_deleg_wait(&delegated_inode);
4571 mnt_drop_write(old_path.mnt);
4573 if (retry_estale(error, lookup_flags))
4574 should_retry = true;
4575 path_put(&new_path);
4578 path_put(&old_path);
4581 should_retry = false;
4582 lookup_flags |= LOOKUP_REVAL;
4589 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4590 int, newdfd, const char __user *, newname)
4592 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4595 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4597 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4600 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4602 int error = may_create(dir, dentry);
4606 if (!dir->i_op->mknod)
4609 return dir->i_op->mknod(dir, dentry,
4610 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4612 EXPORT_SYMBOL(vfs_whiteout);
4614 int readlink_copy(char __user *buffer, int buflen, const char *link)
4616 int len = PTR_ERR(link);
4621 if (len > (unsigned) buflen)
4623 if (copy_to_user(buffer, link, len))
4630 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4631 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4632 * for any given inode is up to filesystem.
4634 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4637 DEFINE_DELAYED_CALL(done);
4638 struct inode *inode = d_inode(dentry);
4639 const char *link = inode->i_link;
4643 link = inode->i_op->get_link(dentry, inode, &done);
4645 return PTR_ERR(link);
4647 res = readlink_copy(buffer, buflen, link);
4648 do_delayed_call(&done);
4653 * vfs_readlink - copy symlink body into userspace buffer
4654 * @dentry: dentry on which to get symbolic link
4655 * @buffer: user memory pointer
4656 * @buflen: size of buffer
4658 * Does not touch atime. That's up to the caller if necessary
4660 * Does not call security hook.
4662 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4664 struct inode *inode = d_inode(dentry);
4666 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4667 if (unlikely(inode->i_op->readlink))
4668 return inode->i_op->readlink(dentry, buffer, buflen);
4670 if (!d_is_symlink(dentry))
4673 spin_lock(&inode->i_lock);
4674 inode->i_opflags |= IOP_DEFAULT_READLINK;
4675 spin_unlock(&inode->i_lock);
4678 return generic_readlink(dentry, buffer, buflen);
4680 EXPORT_SYMBOL(vfs_readlink);
4683 * vfs_get_link - get symlink body
4684 * @dentry: dentry on which to get symbolic link
4685 * @done: caller needs to free returned data with this
4687 * Calls security hook and i_op->get_link() on the supplied inode.
4689 * It does not touch atime. That's up to the caller if necessary.
4691 * Does not work on "special" symlinks like /proc/$$/fd/N
4693 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4695 const char *res = ERR_PTR(-EINVAL);
4696 struct inode *inode = d_inode(dentry);
4698 if (d_is_symlink(dentry)) {
4699 res = ERR_PTR(security_inode_readlink(dentry));
4701 res = inode->i_op->get_link(dentry, inode, done);
4705 EXPORT_SYMBOL(vfs_get_link);
4707 /* get the link contents into pagecache */
4708 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4709 struct delayed_call *callback)
4713 struct address_space *mapping = inode->i_mapping;
4716 page = find_get_page(mapping, 0);
4718 return ERR_PTR(-ECHILD);
4719 if (!PageUptodate(page)) {
4721 return ERR_PTR(-ECHILD);
4724 page = read_mapping_page(mapping, 0, NULL);
4728 set_delayed_call(callback, page_put_link, page);
4729 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4730 kaddr = page_address(page);
4731 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4735 EXPORT_SYMBOL(page_get_link);
4737 void page_put_link(void *arg)
4741 EXPORT_SYMBOL(page_put_link);
4743 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4745 DEFINE_DELAYED_CALL(done);
4746 int res = readlink_copy(buffer, buflen,
4747 page_get_link(dentry, d_inode(dentry),
4749 do_delayed_call(&done);
4752 EXPORT_SYMBOL(page_readlink);
4755 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4757 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4759 struct address_space *mapping = inode->i_mapping;
4763 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4765 flags |= AOP_FLAG_NOFS;
4768 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4769 flags, &page, &fsdata);
4773 memcpy(page_address(page), symname, len-1);
4775 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4782 mark_inode_dirty(inode);
4787 EXPORT_SYMBOL(__page_symlink);
4789 int page_symlink(struct inode *inode, const char *symname, int len)
4791 return __page_symlink(inode, symname, len,
4792 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4794 EXPORT_SYMBOL(page_symlink);
4796 const struct inode_operations page_symlink_inode_operations = {
4797 .get_link = page_get_link,
4799 EXPORT_SYMBOL(page_symlink_inode_operations);