1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 struct filename *tmp;
227 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
228 if (unlikely(!tmp)) {
230 return ERR_PTR(-ENOMEM);
232 tmp->name = (char *)result;
236 return ERR_PTR(-ENAMETOOLONG);
238 memcpy((char *)result->name, filename, len);
240 result->aname = NULL;
242 audit_getname(result);
247 void putname(struct filename *name)
249 BUG_ON(name->refcnt <= 0);
251 if (--name->refcnt > 0)
254 if (name->name != name->iname) {
255 __putname(name->name);
261 static int check_acl(struct inode *inode, int mask)
263 #ifdef CONFIG_FS_POSIX_ACL
264 struct posix_acl *acl;
266 if (mask & MAY_NOT_BLOCK) {
267 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
270 /* no ->get_acl() calls in RCU mode... */
271 if (is_uncached_acl(acl))
273 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
276 acl = get_acl(inode, ACL_TYPE_ACCESS);
280 int error = posix_acl_permission(inode, acl, mask);
281 posix_acl_release(acl);
290 * This does the basic permission checking
292 static int acl_permission_check(struct inode *inode, int mask)
294 unsigned int mode = inode->i_mode;
296 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
299 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
300 int error = check_acl(inode, mask);
301 if (error != -EAGAIN)
305 if (in_group_p(inode->i_gid))
310 * If the DACs are ok we don't need any capability check.
312 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
318 * generic_permission - check for access rights on a Posix-like filesystem
319 * @inode: inode to check access rights for
320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 * Used to check for read/write/execute permissions on a file.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
327 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
328 * request cannot be satisfied (eg. requires blocking or too much complexity).
329 * It would then be called again in ref-walk mode.
331 int generic_permission(struct inode *inode, int mask)
336 * Do the basic permission checks.
338 ret = acl_permission_check(inode, mask);
342 if (S_ISDIR(inode->i_mode)) {
343 /* DACs are overridable for directories */
344 if (!(mask & MAY_WRITE))
345 if (capable_wrt_inode_uidgid(inode,
346 CAP_DAC_READ_SEARCH))
348 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
354 * Searching includes executable on directories, else just read.
356 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
357 if (mask == MAY_READ)
358 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
361 * Read/write DACs are always overridable.
362 * Executable DACs are overridable when there is
363 * at least one exec bit set.
365 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
366 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
371 EXPORT_SYMBOL(generic_permission);
374 * We _really_ want to just do "generic_permission()" without
375 * even looking at the inode->i_op values. So we keep a cache
376 * flag in inode->i_opflags, that says "this has not special
377 * permission function, use the fast case".
379 static inline int do_inode_permission(struct inode *inode, int mask)
381 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
382 if (likely(inode->i_op->permission))
383 return inode->i_op->permission(inode, mask);
385 /* This gets set once for the inode lifetime */
386 spin_lock(&inode->i_lock);
387 inode->i_opflags |= IOP_FASTPERM;
388 spin_unlock(&inode->i_lock);
390 return generic_permission(inode, mask);
394 * __inode_permission - Check for access rights to a given inode
395 * @inode: Inode to check permission on
396 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
398 * Check for read/write/execute permissions on an inode.
400 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
402 * This does not check for a read-only file system. You probably want
403 * inode_permission().
405 int __inode_permission(struct inode *inode, int mask)
409 if (unlikely(mask & MAY_WRITE)) {
411 * Nobody gets write access to an immutable file.
413 if (IS_IMMUTABLE(inode))
417 * Updating mtime will likely cause i_uid and i_gid to be
418 * written back improperly if their true value is unknown
421 if (HAS_UNMAPPED_ID(inode))
425 retval = do_inode_permission(inode, mask);
429 retval = devcgroup_inode_permission(inode, mask);
433 return security_inode_permission(inode, mask);
435 EXPORT_SYMBOL(__inode_permission);
438 * sb_permission - Check superblock-level permissions
439 * @sb: Superblock of inode to check permission on
440 * @inode: Inode to check permission on
441 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
443 * Separate out file-system wide checks from inode-specific permission checks.
445 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
447 if (unlikely(mask & MAY_WRITE)) {
448 umode_t mode = inode->i_mode;
450 /* Nobody gets write access to a read-only fs. */
451 if (sb_rdonly(sb) && (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;
527 } __randomize_layout;
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;
581 struct super_block *sb = mnt->mnt_sb;
583 /* Bind mounts and multi-root filesystems can have disconnected paths */
584 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
587 return is_subdir(path->dentry, mnt->mnt_root);
590 static inline int nd_alloc_stack(struct nameidata *nd)
592 if (likely(nd->depth != EMBEDDED_LEVELS))
594 if (likely(nd->stack != nd->internal))
596 return __nd_alloc_stack(nd);
599 static void drop_links(struct nameidata *nd)
603 struct saved *last = nd->stack + i;
604 do_delayed_call(&last->done);
605 clear_delayed_call(&last->done);
609 static void terminate_walk(struct nameidata *nd)
612 if (!(nd->flags & LOOKUP_RCU)) {
615 for (i = 0; i < nd->depth; i++)
616 path_put(&nd->stack[i].link);
617 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
622 nd->flags &= ~LOOKUP_RCU;
623 if (!(nd->flags & LOOKUP_ROOT))
630 /* path_put is needed afterwards regardless of success or failure */
631 static bool legitimize_path(struct nameidata *nd,
632 struct path *path, unsigned seq)
634 int res = __legitimize_mnt(path->mnt, nd->m_seq);
641 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
645 return !read_seqcount_retry(&path->dentry->d_seq, seq);
648 static bool legitimize_links(struct nameidata *nd)
651 for (i = 0; i < nd->depth; i++) {
652 struct saved *last = nd->stack + i;
653 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
663 * Path walking has 2 modes, rcu-walk and ref-walk (see
664 * Documentation/filesystems/path-lookup.txt). In situations when we can't
665 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
666 * normal reference counts on dentries and vfsmounts to transition to ref-walk
667 * mode. Refcounts are grabbed at the last known good point before rcu-walk
668 * got stuck, so ref-walk may continue from there. If this is not successful
669 * (eg. a seqcount has changed), then failure is returned and it's up to caller
670 * to restart the path walk from the beginning in ref-walk mode.
674 * unlazy_walk - try to switch to ref-walk mode.
675 * @nd: nameidata pathwalk data
676 * Returns: 0 on success, -ECHILD on failure
678 * unlazy_walk attempts to legitimize the current nd->path and nd->root
680 * Must be called from rcu-walk context.
681 * Nothing should touch nameidata between unlazy_walk() failure and
684 static int unlazy_walk(struct nameidata *nd)
686 struct dentry *parent = nd->path.dentry;
688 BUG_ON(!(nd->flags & LOOKUP_RCU));
690 nd->flags &= ~LOOKUP_RCU;
691 if (unlikely(!legitimize_links(nd)))
693 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
695 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
696 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
700 BUG_ON(nd->inode != parent->d_inode);
705 nd->path.dentry = NULL;
707 if (!(nd->flags & LOOKUP_ROOT))
715 * unlazy_child - try to switch to ref-walk mode.
716 * @nd: nameidata pathwalk data
717 * @dentry: child of nd->path.dentry
718 * @seq: seq number to check dentry against
719 * Returns: 0 on success, -ECHILD on failure
721 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
722 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
723 * @nd. Must be called from rcu-walk context.
724 * Nothing should touch nameidata between unlazy_child() failure and
727 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
729 BUG_ON(!(nd->flags & LOOKUP_RCU));
731 nd->flags &= ~LOOKUP_RCU;
732 if (unlikely(!legitimize_links(nd)))
734 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
736 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
740 * We need to move both the parent and the dentry from the RCU domain
741 * to be properly refcounted. And the sequence number in the dentry
742 * validates *both* dentry counters, since we checked the sequence
743 * number of the parent after we got the child sequence number. So we
744 * know the parent must still be valid if the child sequence number is
746 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
748 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
754 * Sequence counts matched. Now make sure that the root is
755 * still valid and get it if required.
757 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
758 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
771 nd->path.dentry = NULL;
775 if (!(nd->flags & LOOKUP_ROOT))
780 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
782 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
783 return dentry->d_op->d_revalidate(dentry, flags);
789 * complete_walk - successful completion of path walk
790 * @nd: pointer nameidata
792 * If we had been in RCU mode, drop out of it and legitimize nd->path.
793 * Revalidate the final result, unless we'd already done that during
794 * the path walk or the filesystem doesn't ask for it. Return 0 on
795 * success, -error on failure. In case of failure caller does not
796 * need to drop nd->path.
798 static int complete_walk(struct nameidata *nd)
800 struct dentry *dentry = nd->path.dentry;
803 if (nd->flags & LOOKUP_RCU) {
804 if (!(nd->flags & LOOKUP_ROOT))
806 if (unlikely(unlazy_walk(nd)))
810 if (likely(!(nd->flags & LOOKUP_JUMPED)))
813 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
816 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
826 static void set_root(struct nameidata *nd)
828 struct fs_struct *fs = current->fs;
830 if (nd->flags & LOOKUP_RCU) {
834 seq = read_seqcount_begin(&fs->seq);
836 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
837 } while (read_seqcount_retry(&fs->seq, seq));
839 get_fs_root(fs, &nd->root);
843 static void path_put_conditional(struct path *path, struct nameidata *nd)
846 if (path->mnt != nd->path.mnt)
850 static inline void path_to_nameidata(const struct path *path,
851 struct nameidata *nd)
853 if (!(nd->flags & LOOKUP_RCU)) {
854 dput(nd->path.dentry);
855 if (nd->path.mnt != path->mnt)
856 mntput(nd->path.mnt);
858 nd->path.mnt = path->mnt;
859 nd->path.dentry = path->dentry;
862 static int nd_jump_root(struct nameidata *nd)
864 if (nd->flags & LOOKUP_RCU) {
868 nd->inode = d->d_inode;
869 nd->seq = nd->root_seq;
870 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
876 nd->inode = nd->path.dentry->d_inode;
878 nd->flags |= LOOKUP_JUMPED;
883 * Helper to directly jump to a known parsed path from ->get_link,
884 * caller must have taken a reference to path beforehand.
886 void nd_jump_link(struct path *path)
888 struct nameidata *nd = current->nameidata;
892 nd->inode = nd->path.dentry->d_inode;
893 nd->flags |= LOOKUP_JUMPED;
896 static inline void put_link(struct nameidata *nd)
898 struct saved *last = nd->stack + --nd->depth;
899 do_delayed_call(&last->done);
900 if (!(nd->flags & LOOKUP_RCU))
901 path_put(&last->link);
904 int sysctl_protected_symlinks __read_mostly = 0;
905 int sysctl_protected_hardlinks __read_mostly = 0;
908 * may_follow_link - Check symlink following for unsafe situations
909 * @nd: nameidata pathwalk data
911 * In the case of the sysctl_protected_symlinks sysctl being enabled,
912 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
913 * in a sticky world-writable directory. This is to protect privileged
914 * processes from failing races against path names that may change out
915 * from under them by way of other users creating malicious symlinks.
916 * It will permit symlinks to be followed only when outside a sticky
917 * world-writable directory, or when the uid of the symlink and follower
918 * match, or when the directory owner matches the symlink's owner.
920 * Returns 0 if following the symlink is allowed, -ve on error.
922 static inline int may_follow_link(struct nameidata *nd)
924 const struct inode *inode;
925 const struct inode *parent;
928 if (!sysctl_protected_symlinks)
931 /* Allowed if owner and follower match. */
932 inode = nd->link_inode;
933 if (uid_eq(current_cred()->fsuid, inode->i_uid))
936 /* Allowed if parent directory not sticky and world-writable. */
938 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
941 /* Allowed if parent directory and link owner match. */
942 puid = parent->i_uid;
943 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
946 if (nd->flags & LOOKUP_RCU)
949 audit_log_link_denied("follow_link", &nd->stack[0].link);
954 * safe_hardlink_source - Check for safe hardlink conditions
955 * @inode: the source inode to hardlink from
957 * Return false if at least one of the following conditions:
958 * - inode is not a regular file
960 * - inode is setgid and group-exec
961 * - access failure for read and write
963 * Otherwise returns true.
965 static bool safe_hardlink_source(struct inode *inode)
967 umode_t mode = inode->i_mode;
969 /* Special files should not get pinned to the filesystem. */
973 /* Setuid files should not get pinned to the filesystem. */
977 /* Executable setgid files should not get pinned to the filesystem. */
978 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
981 /* Hardlinking to unreadable or unwritable sources is dangerous. */
982 if (inode_permission(inode, MAY_READ | MAY_WRITE))
989 * may_linkat - Check permissions for creating a hardlink
990 * @link: the source to hardlink from
992 * Block hardlink when all of:
993 * - sysctl_protected_hardlinks enabled
994 * - fsuid does not match inode
995 * - hardlink source is unsafe (see safe_hardlink_source() above)
996 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
998 * Returns 0 if successful, -ve on error.
1000 static int may_linkat(struct path *link)
1002 struct inode *inode;
1004 if (!sysctl_protected_hardlinks)
1007 inode = link->dentry->d_inode;
1009 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1010 * otherwise, it must be a safe source.
1012 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1015 audit_log_link_denied("linkat", link);
1019 static __always_inline
1020 const char *get_link(struct nameidata *nd)
1022 struct saved *last = nd->stack + nd->depth - 1;
1023 struct dentry *dentry = last->link.dentry;
1024 struct inode *inode = nd->link_inode;
1028 if (!(nd->flags & LOOKUP_RCU)) {
1029 touch_atime(&last->link);
1031 } else if (atime_needs_update_rcu(&last->link, inode)) {
1032 if (unlikely(unlazy_walk(nd)))
1033 return ERR_PTR(-ECHILD);
1034 touch_atime(&last->link);
1037 error = security_inode_follow_link(dentry, inode,
1038 nd->flags & LOOKUP_RCU);
1039 if (unlikely(error))
1040 return ERR_PTR(error);
1042 nd->last_type = LAST_BIND;
1043 res = inode->i_link;
1045 const char * (*get)(struct dentry *, struct inode *,
1046 struct delayed_call *);
1047 get = inode->i_op->get_link;
1048 if (nd->flags & LOOKUP_RCU) {
1049 res = get(NULL, inode, &last->done);
1050 if (res == ERR_PTR(-ECHILD)) {
1051 if (unlikely(unlazy_walk(nd)))
1052 return ERR_PTR(-ECHILD);
1053 res = get(dentry, inode, &last->done);
1056 res = get(dentry, inode, &last->done);
1058 if (IS_ERR_OR_NULL(res))
1064 if (unlikely(nd_jump_root(nd)))
1065 return ERR_PTR(-ECHILD);
1066 while (unlikely(*++res == '/'))
1075 * follow_up - Find the mountpoint of path's vfsmount
1077 * Given a path, find the mountpoint of its source file system.
1078 * Replace @path with the path of the mountpoint in the parent mount.
1081 * Return 1 if we went up a level and 0 if we were already at the
1084 int follow_up(struct path *path)
1086 struct mount *mnt = real_mount(path->mnt);
1087 struct mount *parent;
1088 struct dentry *mountpoint;
1090 read_seqlock_excl(&mount_lock);
1091 parent = mnt->mnt_parent;
1092 if (parent == mnt) {
1093 read_sequnlock_excl(&mount_lock);
1096 mntget(&parent->mnt);
1097 mountpoint = dget(mnt->mnt_mountpoint);
1098 read_sequnlock_excl(&mount_lock);
1100 path->dentry = mountpoint;
1102 path->mnt = &parent->mnt;
1105 EXPORT_SYMBOL(follow_up);
1108 * Perform an automount
1109 * - return -EISDIR to tell follow_managed() to stop and return the path we
1112 static int follow_automount(struct path *path, struct nameidata *nd,
1115 struct vfsmount *mnt;
1118 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1121 /* We don't want to mount if someone's just doing a stat -
1122 * unless they're stat'ing a directory and appended a '/' to
1125 * We do, however, want to mount if someone wants to open or
1126 * create a file of any type under the mountpoint, wants to
1127 * traverse through the mountpoint or wants to open the
1128 * mounted directory. Also, autofs may mark negative dentries
1129 * as being automount points. These will need the attentions
1130 * of the daemon to instantiate them before they can be used.
1132 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1133 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1134 path->dentry->d_inode)
1137 nd->total_link_count++;
1138 if (nd->total_link_count >= 40)
1141 mnt = path->dentry->d_op->d_automount(path);
1144 * The filesystem is allowed to return -EISDIR here to indicate
1145 * it doesn't want to automount. For instance, autofs would do
1146 * this so that its userspace daemon can mount on this dentry.
1148 * However, we can only permit this if it's a terminal point in
1149 * the path being looked up; if it wasn't then the remainder of
1150 * the path is inaccessible and we should say so.
1152 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1154 return PTR_ERR(mnt);
1157 if (!mnt) /* mount collision */
1160 if (!*need_mntput) {
1161 /* lock_mount() may release path->mnt on error */
1163 *need_mntput = true;
1165 err = finish_automount(mnt, path);
1169 /* Someone else made a mount here whilst we were busy */
1174 path->dentry = dget(mnt->mnt_root);
1183 * Handle a dentry that is managed in some way.
1184 * - Flagged for transit management (autofs)
1185 * - Flagged as mountpoint
1186 * - Flagged as automount point
1188 * This may only be called in refwalk mode.
1190 * Serialization is taken care of in namespace.c
1192 static int follow_managed(struct path *path, struct nameidata *nd)
1194 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1196 bool need_mntput = false;
1199 /* Given that we're not holding a lock here, we retain the value in a
1200 * local variable for each dentry as we look at it so that we don't see
1201 * the components of that value change under us */
1202 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1203 managed &= DCACHE_MANAGED_DENTRY,
1204 unlikely(managed != 0)) {
1205 /* Allow the filesystem to manage the transit without i_mutex
1207 if (managed & DCACHE_MANAGE_TRANSIT) {
1208 BUG_ON(!path->dentry->d_op);
1209 BUG_ON(!path->dentry->d_op->d_manage);
1210 ret = path->dentry->d_op->d_manage(path, false);
1215 /* Transit to a mounted filesystem. */
1216 if (managed & DCACHE_MOUNTED) {
1217 struct vfsmount *mounted = lookup_mnt(path);
1222 path->mnt = mounted;
1223 path->dentry = dget(mounted->mnt_root);
1228 /* Something is mounted on this dentry in another
1229 * namespace and/or whatever was mounted there in this
1230 * namespace got unmounted before lookup_mnt() could
1234 /* Handle an automount point */
1235 if (managed & DCACHE_NEED_AUTOMOUNT) {
1236 ret = follow_automount(path, nd, &need_mntput);
1242 /* We didn't change the current path point */
1246 if (need_mntput && path->mnt == mnt)
1248 if (ret == -EISDIR || !ret)
1251 nd->flags |= LOOKUP_JUMPED;
1252 if (unlikely(ret < 0))
1253 path_put_conditional(path, nd);
1257 int follow_down_one(struct path *path)
1259 struct vfsmount *mounted;
1261 mounted = lookup_mnt(path);
1265 path->mnt = mounted;
1266 path->dentry = dget(mounted->mnt_root);
1271 EXPORT_SYMBOL(follow_down_one);
1273 static inline int managed_dentry_rcu(const struct path *path)
1275 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1276 path->dentry->d_op->d_manage(path, true) : 0;
1280 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1281 * we meet a managed dentry that would need blocking.
1283 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1284 struct inode **inode, unsigned *seqp)
1287 struct mount *mounted;
1289 * Don't forget we might have a non-mountpoint managed dentry
1290 * that wants to block transit.
1292 switch (managed_dentry_rcu(path)) {
1302 if (!d_mountpoint(path->dentry))
1303 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1305 mounted = __lookup_mnt(path->mnt, path->dentry);
1308 path->mnt = &mounted->mnt;
1309 path->dentry = mounted->mnt.mnt_root;
1310 nd->flags |= LOOKUP_JUMPED;
1311 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1313 * Update the inode too. We don't need to re-check the
1314 * dentry sequence number here after this d_inode read,
1315 * because a mount-point is always pinned.
1317 *inode = path->dentry->d_inode;
1319 return !read_seqretry(&mount_lock, nd->m_seq) &&
1320 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1323 static int follow_dotdot_rcu(struct nameidata *nd)
1325 struct inode *inode = nd->inode;
1328 if (path_equal(&nd->path, &nd->root))
1330 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1331 struct dentry *old = nd->path.dentry;
1332 struct dentry *parent = old->d_parent;
1335 inode = parent->d_inode;
1336 seq = read_seqcount_begin(&parent->d_seq);
1337 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1339 nd->path.dentry = parent;
1341 if (unlikely(!path_connected(&nd->path)))
1345 struct mount *mnt = real_mount(nd->path.mnt);
1346 struct mount *mparent = mnt->mnt_parent;
1347 struct dentry *mountpoint = mnt->mnt_mountpoint;
1348 struct inode *inode2 = mountpoint->d_inode;
1349 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1350 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1352 if (&mparent->mnt == nd->path.mnt)
1354 /* we know that mountpoint was pinned */
1355 nd->path.dentry = mountpoint;
1356 nd->path.mnt = &mparent->mnt;
1361 while (unlikely(d_mountpoint(nd->path.dentry))) {
1362 struct mount *mounted;
1363 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1364 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1368 nd->path.mnt = &mounted->mnt;
1369 nd->path.dentry = mounted->mnt.mnt_root;
1370 inode = nd->path.dentry->d_inode;
1371 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1378 * Follow down to the covering mount currently visible to userspace. At each
1379 * point, the filesystem owning that dentry may be queried as to whether the
1380 * caller is permitted to proceed or not.
1382 int follow_down(struct path *path)
1387 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1388 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1389 /* Allow the filesystem to manage the transit without i_mutex
1392 * We indicate to the filesystem if someone is trying to mount
1393 * something here. This gives autofs the chance to deny anyone
1394 * other than its daemon the right to mount on its
1397 * The filesystem may sleep at this point.
1399 if (managed & DCACHE_MANAGE_TRANSIT) {
1400 BUG_ON(!path->dentry->d_op);
1401 BUG_ON(!path->dentry->d_op->d_manage);
1402 ret = path->dentry->d_op->d_manage(path, false);
1404 return ret == -EISDIR ? 0 : ret;
1407 /* Transit to a mounted filesystem. */
1408 if (managed & DCACHE_MOUNTED) {
1409 struct vfsmount *mounted = lookup_mnt(path);
1414 path->mnt = mounted;
1415 path->dentry = dget(mounted->mnt_root);
1419 /* Don't handle automount points here */
1424 EXPORT_SYMBOL(follow_down);
1427 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1429 static void follow_mount(struct path *path)
1431 while (d_mountpoint(path->dentry)) {
1432 struct vfsmount *mounted = lookup_mnt(path);
1437 path->mnt = mounted;
1438 path->dentry = dget(mounted->mnt_root);
1442 static int path_parent_directory(struct path *path)
1444 struct dentry *old = path->dentry;
1445 /* rare case of legitimate dget_parent()... */
1446 path->dentry = dget_parent(path->dentry);
1448 if (unlikely(!path_connected(path)))
1453 static int follow_dotdot(struct nameidata *nd)
1456 if (nd->path.dentry == nd->root.dentry &&
1457 nd->path.mnt == nd->root.mnt) {
1460 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1461 int ret = path_parent_directory(&nd->path);
1466 if (!follow_up(&nd->path))
1469 follow_mount(&nd->path);
1470 nd->inode = nd->path.dentry->d_inode;
1475 * This looks up the name in dcache and possibly revalidates the found dentry.
1476 * NULL is returned if the dentry does not exist in the cache.
1478 static struct dentry *lookup_dcache(const struct qstr *name,
1482 struct dentry *dentry = d_lookup(dir, name);
1484 int error = d_revalidate(dentry, flags);
1485 if (unlikely(error <= 0)) {
1487 d_invalidate(dentry);
1489 return ERR_PTR(error);
1496 * Call i_op->lookup on the dentry. The dentry must be negative and
1499 * dir->d_inode->i_mutex must be held
1501 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1506 /* Don't create child dentry for a dead directory. */
1507 if (unlikely(IS_DEADDIR(dir))) {
1509 return ERR_PTR(-ENOENT);
1512 old = dir->i_op->lookup(dir, dentry, flags);
1513 if (unlikely(old)) {
1520 static struct dentry *__lookup_hash(const struct qstr *name,
1521 struct dentry *base, unsigned int flags)
1523 struct dentry *dentry = lookup_dcache(name, base, flags);
1528 dentry = d_alloc(base, name);
1529 if (unlikely(!dentry))
1530 return ERR_PTR(-ENOMEM);
1532 return lookup_real(base->d_inode, dentry, flags);
1535 static int lookup_fast(struct nameidata *nd,
1536 struct path *path, struct inode **inode,
1539 struct vfsmount *mnt = nd->path.mnt;
1540 struct dentry *dentry, *parent = nd->path.dentry;
1545 * Rename seqlock is not required here because in the off chance
1546 * of a false negative due to a concurrent rename, the caller is
1547 * going to fall back to non-racy lookup.
1549 if (nd->flags & LOOKUP_RCU) {
1552 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1553 if (unlikely(!dentry)) {
1554 if (unlazy_walk(nd))
1560 * This sequence count validates that the inode matches
1561 * the dentry name information from lookup.
1563 *inode = d_backing_inode(dentry);
1564 negative = d_is_negative(dentry);
1565 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1569 * This sequence count validates that the parent had no
1570 * changes while we did the lookup of the dentry above.
1572 * The memory barrier in read_seqcount_begin of child is
1573 * enough, we can use __read_seqcount_retry here.
1575 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1579 status = d_revalidate(dentry, nd->flags);
1580 if (likely(status > 0)) {
1582 * Note: do negative dentry check after revalidation in
1583 * case that drops it.
1585 if (unlikely(negative))
1588 path->dentry = dentry;
1589 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1592 if (unlazy_child(nd, dentry, seq))
1594 if (unlikely(status == -ECHILD))
1595 /* we'd been told to redo it in non-rcu mode */
1596 status = d_revalidate(dentry, nd->flags);
1598 dentry = __d_lookup(parent, &nd->last);
1599 if (unlikely(!dentry))
1601 status = d_revalidate(dentry, nd->flags);
1603 if (unlikely(status <= 0)) {
1605 d_invalidate(dentry);
1609 if (unlikely(d_is_negative(dentry))) {
1615 path->dentry = dentry;
1616 err = follow_managed(path, nd);
1617 if (likely(err > 0))
1618 *inode = d_backing_inode(path->dentry);
1622 /* Fast lookup failed, do it the slow way */
1623 static struct dentry *lookup_slow(const struct qstr *name,
1627 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1628 struct inode *inode = dir->d_inode;
1629 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1631 inode_lock_shared(inode);
1632 /* Don't go there if it's already dead */
1633 if (unlikely(IS_DEADDIR(inode)))
1636 dentry = d_alloc_parallel(dir, name, &wq);
1639 if (unlikely(!d_in_lookup(dentry))) {
1640 if (!(flags & LOOKUP_NO_REVAL)) {
1641 int error = d_revalidate(dentry, flags);
1642 if (unlikely(error <= 0)) {
1644 d_invalidate(dentry);
1649 dentry = ERR_PTR(error);
1653 old = inode->i_op->lookup(inode, dentry, flags);
1654 d_lookup_done(dentry);
1655 if (unlikely(old)) {
1661 inode_unlock_shared(inode);
1665 static inline int may_lookup(struct nameidata *nd)
1667 if (nd->flags & LOOKUP_RCU) {
1668 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1671 if (unlazy_walk(nd))
1674 return inode_permission(nd->inode, MAY_EXEC);
1677 static inline int handle_dots(struct nameidata *nd, int type)
1679 if (type == LAST_DOTDOT) {
1682 if (nd->flags & LOOKUP_RCU) {
1683 return follow_dotdot_rcu(nd);
1685 return follow_dotdot(nd);
1690 static int pick_link(struct nameidata *nd, struct path *link,
1691 struct inode *inode, unsigned seq)
1695 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1696 path_to_nameidata(link, nd);
1699 if (!(nd->flags & LOOKUP_RCU)) {
1700 if (link->mnt == nd->path.mnt)
1703 error = nd_alloc_stack(nd);
1704 if (unlikely(error)) {
1705 if (error == -ECHILD) {
1706 if (unlikely(!legitimize_path(nd, link, seq))) {
1709 nd->flags &= ~LOOKUP_RCU;
1710 nd->path.mnt = NULL;
1711 nd->path.dentry = NULL;
1712 if (!(nd->flags & LOOKUP_ROOT))
1713 nd->root.mnt = NULL;
1715 } else if (likely(unlazy_walk(nd)) == 0)
1716 error = nd_alloc_stack(nd);
1724 last = nd->stack + nd->depth++;
1726 clear_delayed_call(&last->done);
1727 nd->link_inode = inode;
1732 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1735 * Do we need to follow links? We _really_ want to be able
1736 * to do this check without having to look at inode->i_op,
1737 * so we keep a cache of "no, this doesn't need follow_link"
1738 * for the common case.
1740 static inline int step_into(struct nameidata *nd, struct path *path,
1741 int flags, struct inode *inode, unsigned seq)
1743 if (!(flags & WALK_MORE) && nd->depth)
1745 if (likely(!d_is_symlink(path->dentry)) ||
1746 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1747 /* not a symlink or should not follow */
1748 path_to_nameidata(path, nd);
1753 /* make sure that d_is_symlink above matches inode */
1754 if (nd->flags & LOOKUP_RCU) {
1755 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1758 return pick_link(nd, path, inode, seq);
1761 static int walk_component(struct nameidata *nd, int flags)
1764 struct inode *inode;
1768 * "." and ".." are special - ".." especially so because it has
1769 * to be able to know about the current root directory and
1770 * parent relationships.
1772 if (unlikely(nd->last_type != LAST_NORM)) {
1773 err = handle_dots(nd, nd->last_type);
1774 if (!(flags & WALK_MORE) && nd->depth)
1778 err = lookup_fast(nd, &path, &inode, &seq);
1779 if (unlikely(err <= 0)) {
1782 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1784 if (IS_ERR(path.dentry))
1785 return PTR_ERR(path.dentry);
1787 path.mnt = nd->path.mnt;
1788 err = follow_managed(&path, nd);
1789 if (unlikely(err < 0))
1792 if (unlikely(d_is_negative(path.dentry))) {
1793 path_to_nameidata(&path, nd);
1797 seq = 0; /* we are already out of RCU mode */
1798 inode = d_backing_inode(path.dentry);
1801 return step_into(nd, &path, flags, inode, seq);
1805 * We can do the critical dentry name comparison and hashing
1806 * operations one word at a time, but we are limited to:
1808 * - Architectures with fast unaligned word accesses. We could
1809 * do a "get_unaligned()" if this helps and is sufficiently
1812 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1813 * do not trap on the (extremely unlikely) case of a page
1814 * crossing operation.
1816 * - Furthermore, we need an efficient 64-bit compile for the
1817 * 64-bit case in order to generate the "number of bytes in
1818 * the final mask". Again, that could be replaced with a
1819 * efficient population count instruction or similar.
1821 #ifdef CONFIG_DCACHE_WORD_ACCESS
1823 #include <asm/word-at-a-time.h>
1827 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1829 #elif defined(CONFIG_64BIT)
1831 * Register pressure in the mixing function is an issue, particularly
1832 * on 32-bit x86, but almost any function requires one state value and
1833 * one temporary. Instead, use a function designed for two state values
1834 * and no temporaries.
1836 * This function cannot create a collision in only two iterations, so
1837 * we have two iterations to achieve avalanche. In those two iterations,
1838 * we have six layers of mixing, which is enough to spread one bit's
1839 * influence out to 2^6 = 64 state bits.
1841 * Rotate constants are scored by considering either 64 one-bit input
1842 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1843 * probability of that delta causing a change to each of the 128 output
1844 * bits, using a sample of random initial states.
1846 * The Shannon entropy of the computed probabilities is then summed
1847 * to produce a score. Ideally, any input change has a 50% chance of
1848 * toggling any given output bit.
1850 * Mixing scores (in bits) for (12,45):
1851 * Input delta: 1-bit 2-bit
1852 * 1 round: 713.3 42542.6
1853 * 2 rounds: 2753.7 140389.8
1854 * 3 rounds: 5954.1 233458.2
1855 * 4 rounds: 7862.6 256672.2
1856 * Perfect: 8192 258048
1857 * (64*128) (64*63/2 * 128)
1859 #define HASH_MIX(x, y, a) \
1861 y ^= x, x = rol64(x,12),\
1862 x += y, y = rol64(y,45),\
1866 * Fold two longs into one 32-bit hash value. This must be fast, but
1867 * latency isn't quite as critical, as there is a fair bit of additional
1868 * work done before the hash value is used.
1870 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1872 y ^= x * GOLDEN_RATIO_64;
1873 y *= GOLDEN_RATIO_64;
1877 #else /* 32-bit case */
1880 * Mixing scores (in bits) for (7,20):
1881 * Input delta: 1-bit 2-bit
1882 * 1 round: 330.3 9201.6
1883 * 2 rounds: 1246.4 25475.4
1884 * 3 rounds: 1907.1 31295.1
1885 * 4 rounds: 2042.3 31718.6
1886 * Perfect: 2048 31744
1887 * (32*64) (32*31/2 * 64)
1889 #define HASH_MIX(x, y, a) \
1891 y ^= x, x = rol32(x, 7),\
1892 x += y, y = rol32(y,20),\
1895 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1897 /* Use arch-optimized multiply if one exists */
1898 return __hash_32(y ^ __hash_32(x));
1904 * Return the hash of a string of known length. This is carfully
1905 * designed to match hash_name(), which is the more critical function.
1906 * In particular, we must end by hashing a final word containing 0..7
1907 * payload bytes, to match the way that hash_name() iterates until it
1908 * finds the delimiter after the name.
1910 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1912 unsigned long a, x = 0, y = (unsigned long)salt;
1917 a = load_unaligned_zeropad(name);
1918 if (len < sizeof(unsigned long))
1921 name += sizeof(unsigned long);
1922 len -= sizeof(unsigned long);
1924 x ^= a & bytemask_from_count(len);
1926 return fold_hash(x, y);
1928 EXPORT_SYMBOL(full_name_hash);
1930 /* Return the "hash_len" (hash and length) of a null-terminated string */
1931 u64 hashlen_string(const void *salt, const char *name)
1933 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1934 unsigned long adata, mask, len;
1935 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1942 len += sizeof(unsigned long);
1944 a = load_unaligned_zeropad(name+len);
1945 } while (!has_zero(a, &adata, &constants));
1947 adata = prep_zero_mask(a, adata, &constants);
1948 mask = create_zero_mask(adata);
1949 x ^= a & zero_bytemask(mask);
1951 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1953 EXPORT_SYMBOL(hashlen_string);
1956 * Calculate the length and hash of the path component, and
1957 * return the "hash_len" as the result.
1959 static inline u64 hash_name(const void *salt, const char *name)
1961 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1962 unsigned long adata, bdata, mask, len;
1963 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1970 len += sizeof(unsigned long);
1972 a = load_unaligned_zeropad(name+len);
1973 b = a ^ REPEAT_BYTE('/');
1974 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1976 adata = prep_zero_mask(a, adata, &constants);
1977 bdata = prep_zero_mask(b, bdata, &constants);
1978 mask = create_zero_mask(adata | bdata);
1979 x ^= a & zero_bytemask(mask);
1981 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1984 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1986 /* Return the hash of a string of known length */
1987 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1989 unsigned long hash = init_name_hash(salt);
1991 hash = partial_name_hash((unsigned char)*name++, hash);
1992 return end_name_hash(hash);
1994 EXPORT_SYMBOL(full_name_hash);
1996 /* Return the "hash_len" (hash and length) of a null-terminated string */
1997 u64 hashlen_string(const void *salt, const char *name)
1999 unsigned long hash = init_name_hash(salt);
2000 unsigned long len = 0, c;
2002 c = (unsigned char)*name;
2005 hash = partial_name_hash(c, hash);
2006 c = (unsigned char)name[len];
2008 return hashlen_create(end_name_hash(hash), len);
2010 EXPORT_SYMBOL(hashlen_string);
2013 * We know there's a real path component here of at least
2016 static inline u64 hash_name(const void *salt, const char *name)
2018 unsigned long hash = init_name_hash(salt);
2019 unsigned long len = 0, c;
2021 c = (unsigned char)*name;
2024 hash = partial_name_hash(c, hash);
2025 c = (unsigned char)name[len];
2026 } while (c && c != '/');
2027 return hashlen_create(end_name_hash(hash), len);
2034 * This is the basic name resolution function, turning a pathname into
2035 * the final dentry. We expect 'base' to be positive and a directory.
2037 * Returns 0 and nd will have valid dentry and mnt on success.
2038 * Returns error and drops reference to input namei data on failure.
2040 static int link_path_walk(const char *name, struct nameidata *nd)
2049 /* At this point we know we have a real path component. */
2054 err = may_lookup(nd);
2058 hash_len = hash_name(nd->path.dentry, name);
2061 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2063 if (name[1] == '.') {
2065 nd->flags |= LOOKUP_JUMPED;
2071 if (likely(type == LAST_NORM)) {
2072 struct dentry *parent = nd->path.dentry;
2073 nd->flags &= ~LOOKUP_JUMPED;
2074 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2075 struct qstr this = { { .hash_len = hash_len }, .name = name };
2076 err = parent->d_op->d_hash(parent, &this);
2079 hash_len = this.hash_len;
2084 nd->last.hash_len = hash_len;
2085 nd->last.name = name;
2086 nd->last_type = type;
2088 name += hashlen_len(hash_len);
2092 * If it wasn't NUL, we know it was '/'. Skip that
2093 * slash, and continue until no more slashes.
2097 } while (unlikely(*name == '/'));
2098 if (unlikely(!*name)) {
2100 /* pathname body, done */
2103 name = nd->stack[nd->depth - 1].name;
2104 /* trailing symlink, done */
2107 /* last component of nested symlink */
2108 err = walk_component(nd, WALK_FOLLOW);
2110 /* not the last component */
2111 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2117 const char *s = get_link(nd);
2126 nd->stack[nd->depth - 1].name = name;
2131 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2132 if (nd->flags & LOOKUP_RCU) {
2133 if (unlazy_walk(nd))
2141 static const char *path_init(struct nameidata *nd, unsigned flags)
2143 const char *s = nd->name->name;
2146 flags &= ~LOOKUP_RCU;
2148 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2149 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2151 if (flags & LOOKUP_ROOT) {
2152 struct dentry *root = nd->root.dentry;
2153 struct inode *inode = root->d_inode;
2154 if (*s && unlikely(!d_can_lookup(root)))
2155 return ERR_PTR(-ENOTDIR);
2156 nd->path = nd->root;
2158 if (flags & LOOKUP_RCU) {
2160 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2161 nd->root_seq = nd->seq;
2162 nd->m_seq = read_seqbegin(&mount_lock);
2164 path_get(&nd->path);
2169 nd->root.mnt = NULL;
2170 nd->path.mnt = NULL;
2171 nd->path.dentry = NULL;
2173 nd->m_seq = read_seqbegin(&mount_lock);
2175 if (flags & LOOKUP_RCU)
2178 if (likely(!nd_jump_root(nd)))
2180 nd->root.mnt = NULL;
2182 return ERR_PTR(-ECHILD);
2183 } else if (nd->dfd == AT_FDCWD) {
2184 if (flags & LOOKUP_RCU) {
2185 struct fs_struct *fs = current->fs;
2191 seq = read_seqcount_begin(&fs->seq);
2193 nd->inode = nd->path.dentry->d_inode;
2194 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2195 } while (read_seqcount_retry(&fs->seq, seq));
2197 get_fs_pwd(current->fs, &nd->path);
2198 nd->inode = nd->path.dentry->d_inode;
2202 /* Caller must check execute permissions on the starting path component */
2203 struct fd f = fdget_raw(nd->dfd);
2204 struct dentry *dentry;
2207 return ERR_PTR(-EBADF);
2209 dentry = f.file->f_path.dentry;
2212 if (!d_can_lookup(dentry)) {
2214 return ERR_PTR(-ENOTDIR);
2218 nd->path = f.file->f_path;
2219 if (flags & LOOKUP_RCU) {
2221 nd->inode = nd->path.dentry->d_inode;
2222 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2224 path_get(&nd->path);
2225 nd->inode = nd->path.dentry->d_inode;
2232 static const char *trailing_symlink(struct nameidata *nd)
2235 int error = may_follow_link(nd);
2236 if (unlikely(error))
2237 return ERR_PTR(error);
2238 nd->flags |= LOOKUP_PARENT;
2239 nd->stack[0].name = NULL;
2244 static inline int lookup_last(struct nameidata *nd)
2246 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2247 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2249 nd->flags &= ~LOOKUP_PARENT;
2250 return walk_component(nd, 0);
2253 static int handle_lookup_down(struct nameidata *nd)
2255 struct path path = nd->path;
2256 struct inode *inode = nd->inode;
2257 unsigned seq = nd->seq;
2260 if (nd->flags & LOOKUP_RCU) {
2262 * don't bother with unlazy_walk on failure - we are
2263 * at the very beginning of walk, so we lose nothing
2264 * if we simply redo everything in non-RCU mode
2266 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2270 err = follow_managed(&path, nd);
2271 if (unlikely(err < 0))
2273 inode = d_backing_inode(path.dentry);
2276 path_to_nameidata(&path, nd);
2282 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2283 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2285 const char *s = path_init(nd, flags);
2291 if (unlikely(flags & LOOKUP_DOWN)) {
2292 err = handle_lookup_down(nd);
2293 if (unlikely(err < 0)) {
2299 while (!(err = link_path_walk(s, nd))
2300 && ((err = lookup_last(nd)) > 0)) {
2301 s = trailing_symlink(nd);
2308 err = complete_walk(nd);
2310 if (!err && nd->flags & LOOKUP_DIRECTORY)
2311 if (!d_can_lookup(nd->path.dentry))
2315 nd->path.mnt = NULL;
2316 nd->path.dentry = NULL;
2322 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2323 struct path *path, struct path *root)
2326 struct nameidata nd;
2328 return PTR_ERR(name);
2329 if (unlikely(root)) {
2331 flags |= LOOKUP_ROOT;
2333 set_nameidata(&nd, dfd, name);
2334 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2335 if (unlikely(retval == -ECHILD))
2336 retval = path_lookupat(&nd, flags, path);
2337 if (unlikely(retval == -ESTALE))
2338 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2340 if (likely(!retval))
2341 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2342 restore_nameidata();
2347 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2348 static int path_parentat(struct nameidata *nd, unsigned flags,
2349 struct path *parent)
2351 const char *s = path_init(nd, flags);
2355 err = link_path_walk(s, nd);
2357 err = complete_walk(nd);
2360 nd->path.mnt = NULL;
2361 nd->path.dentry = NULL;
2367 static struct filename *filename_parentat(int dfd, struct filename *name,
2368 unsigned int flags, struct path *parent,
2369 struct qstr *last, int *type)
2372 struct nameidata nd;
2376 set_nameidata(&nd, dfd, name);
2377 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2378 if (unlikely(retval == -ECHILD))
2379 retval = path_parentat(&nd, flags, parent);
2380 if (unlikely(retval == -ESTALE))
2381 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2382 if (likely(!retval)) {
2384 *type = nd.last_type;
2385 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2388 name = ERR_PTR(retval);
2390 restore_nameidata();
2394 /* does lookup, returns the object with parent locked */
2395 struct dentry *kern_path_locked(const char *name, struct path *path)
2397 struct filename *filename;
2402 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2404 if (IS_ERR(filename))
2405 return ERR_CAST(filename);
2406 if (unlikely(type != LAST_NORM)) {
2409 return ERR_PTR(-EINVAL);
2411 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2412 d = __lookup_hash(&last, path->dentry, 0);
2414 inode_unlock(path->dentry->d_inode);
2421 int kern_path(const char *name, unsigned int flags, struct path *path)
2423 return filename_lookup(AT_FDCWD, getname_kernel(name),
2426 EXPORT_SYMBOL(kern_path);
2429 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2430 * @dentry: pointer to dentry of the base directory
2431 * @mnt: pointer to vfs mount of the base directory
2432 * @name: pointer to file name
2433 * @flags: lookup flags
2434 * @path: pointer to struct path to fill
2436 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2437 const char *name, unsigned int flags,
2440 struct path root = {.mnt = mnt, .dentry = dentry};
2441 /* the first argument of filename_lookup() is ignored with root */
2442 return filename_lookup(AT_FDCWD, getname_kernel(name),
2443 flags , path, &root);
2445 EXPORT_SYMBOL(vfs_path_lookup);
2448 * lookup_one_len - filesystem helper to lookup single pathname component
2449 * @name: pathname component to lookup
2450 * @base: base directory to lookup from
2451 * @len: maximum length @len should be interpreted to
2453 * Note that this routine is purely a helper for filesystem usage and should
2454 * not be called by generic code.
2456 * The caller must hold base->i_mutex.
2458 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2464 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2468 this.hash = full_name_hash(base, name, len);
2470 return ERR_PTR(-EACCES);
2472 if (unlikely(name[0] == '.')) {
2473 if (len < 2 || (len == 2 && name[1] == '.'))
2474 return ERR_PTR(-EACCES);
2478 c = *(const unsigned char *)name++;
2479 if (c == '/' || c == '\0')
2480 return ERR_PTR(-EACCES);
2483 * See if the low-level filesystem might want
2484 * to use its own hash..
2486 if (base->d_flags & DCACHE_OP_HASH) {
2487 int err = base->d_op->d_hash(base, &this);
2489 return ERR_PTR(err);
2492 err = inode_permission(base->d_inode, MAY_EXEC);
2494 return ERR_PTR(err);
2496 return __lookup_hash(&this, base, 0);
2498 EXPORT_SYMBOL(lookup_one_len);
2501 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2502 * @name: pathname component to lookup
2503 * @base: base directory to lookup from
2504 * @len: maximum length @len should be interpreted to
2506 * Note that this routine is purely a helper for filesystem usage and should
2507 * not be called by generic code.
2509 * Unlike lookup_one_len, it should be called without the parent
2510 * i_mutex held, and will take the i_mutex itself if necessary.
2512 struct dentry *lookup_one_len_unlocked(const char *name,
2513 struct dentry *base, int len)
2522 this.hash = full_name_hash(base, name, len);
2524 return ERR_PTR(-EACCES);
2526 if (unlikely(name[0] == '.')) {
2527 if (len < 2 || (len == 2 && name[1] == '.'))
2528 return ERR_PTR(-EACCES);
2532 c = *(const unsigned char *)name++;
2533 if (c == '/' || c == '\0')
2534 return ERR_PTR(-EACCES);
2537 * See if the low-level filesystem might want
2538 * to use its own hash..
2540 if (base->d_flags & DCACHE_OP_HASH) {
2541 int err = base->d_op->d_hash(base, &this);
2543 return ERR_PTR(err);
2546 err = inode_permission(base->d_inode, MAY_EXEC);
2548 return ERR_PTR(err);
2550 ret = lookup_dcache(&this, base, 0);
2552 ret = lookup_slow(&this, base, 0);
2555 EXPORT_SYMBOL(lookup_one_len_unlocked);
2557 #ifdef CONFIG_UNIX98_PTYS
2558 int path_pts(struct path *path)
2560 /* Find something mounted on "pts" in the same directory as
2563 struct dentry *child, *parent;
2567 ret = path_parent_directory(path);
2571 parent = path->dentry;
2574 child = d_hash_and_lookup(parent, &this);
2578 path->dentry = child;
2585 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2586 struct path *path, int *empty)
2588 return filename_lookup(dfd, getname_flags(name, flags, empty),
2591 EXPORT_SYMBOL(user_path_at_empty);
2594 * mountpoint_last - look up last component for umount
2595 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2597 * This is a special lookup_last function just for umount. In this case, we
2598 * need to resolve the path without doing any revalidation.
2600 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2601 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2602 * in almost all cases, this lookup will be served out of the dcache. The only
2603 * cases where it won't are if nd->last refers to a symlink or the path is
2604 * bogus and it doesn't exist.
2607 * -error: if there was an error during lookup. This includes -ENOENT if the
2608 * lookup found a negative dentry.
2610 * 0: if we successfully resolved nd->last and found it to not to be a
2611 * symlink that needs to be followed.
2613 * 1: if we successfully resolved nd->last and found it to be a symlink
2614 * that needs to be followed.
2617 mountpoint_last(struct nameidata *nd)
2620 struct dentry *dir = nd->path.dentry;
2623 /* If we're in rcuwalk, drop out of it to handle last component */
2624 if (nd->flags & LOOKUP_RCU) {
2625 if (unlazy_walk(nd))
2629 nd->flags &= ~LOOKUP_PARENT;
2631 if (unlikely(nd->last_type != LAST_NORM)) {
2632 error = handle_dots(nd, nd->last_type);
2635 path.dentry = dget(nd->path.dentry);
2637 path.dentry = d_lookup(dir, &nd->last);
2640 * No cached dentry. Mounted dentries are pinned in the
2641 * cache, so that means that this dentry is probably
2642 * a symlink or the path doesn't actually point
2643 * to a mounted dentry.
2645 path.dentry = lookup_slow(&nd->last, dir,
2646 nd->flags | LOOKUP_NO_REVAL);
2647 if (IS_ERR(path.dentry))
2648 return PTR_ERR(path.dentry);
2651 if (d_is_negative(path.dentry)) {
2655 path.mnt = nd->path.mnt;
2656 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2660 * path_mountpoint - look up a path to be umounted
2661 * @nd: lookup context
2662 * @flags: lookup flags
2663 * @path: pointer to container for result
2665 * Look up the given name, but don't attempt to revalidate the last component.
2666 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2669 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2671 const char *s = path_init(nd, flags);
2675 while (!(err = link_path_walk(s, nd)) &&
2676 (err = mountpoint_last(nd)) > 0) {
2677 s = trailing_symlink(nd);
2685 nd->path.mnt = NULL;
2686 nd->path.dentry = NULL;
2694 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2697 struct nameidata nd;
2700 return PTR_ERR(name);
2701 set_nameidata(&nd, dfd, name);
2702 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2703 if (unlikely(error == -ECHILD))
2704 error = path_mountpoint(&nd, flags, path);
2705 if (unlikely(error == -ESTALE))
2706 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2708 audit_inode(name, path->dentry, 0);
2709 restore_nameidata();
2715 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2716 * @dfd: directory file descriptor
2717 * @name: pathname from userland
2718 * @flags: lookup flags
2719 * @path: pointer to container to hold result
2721 * A umount is a special case for path walking. We're not actually interested
2722 * in the inode in this situation, and ESTALE errors can be a problem. We
2723 * simply want track down the dentry and vfsmount attached at the mountpoint
2724 * and avoid revalidating the last component.
2726 * Returns 0 and populates "path" on success.
2729 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2732 return filename_mountpoint(dfd, getname(name), path, flags);
2736 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2739 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2741 EXPORT_SYMBOL(kern_path_mountpoint);
2743 int __check_sticky(struct inode *dir, struct inode *inode)
2745 kuid_t fsuid = current_fsuid();
2747 if (uid_eq(inode->i_uid, fsuid))
2749 if (uid_eq(dir->i_uid, fsuid))
2751 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2753 EXPORT_SYMBOL(__check_sticky);
2756 * Check whether we can remove a link victim from directory dir, check
2757 * whether the type of victim is right.
2758 * 1. We can't do it if dir is read-only (done in permission())
2759 * 2. We should have write and exec permissions on dir
2760 * 3. We can't remove anything from append-only dir
2761 * 4. We can't do anything with immutable dir (done in permission())
2762 * 5. If the sticky bit on dir is set we should either
2763 * a. be owner of dir, or
2764 * b. be owner of victim, or
2765 * c. have CAP_FOWNER capability
2766 * 6. If the victim is append-only or immutable we can't do antyhing with
2767 * links pointing to it.
2768 * 7. If the victim has an unknown uid or gid we can't change the inode.
2769 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2770 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2771 * 10. We can't remove a root or mountpoint.
2772 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2773 * nfs_async_unlink().
2775 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2777 struct inode *inode = d_backing_inode(victim);
2780 if (d_is_negative(victim))
2784 BUG_ON(victim->d_parent->d_inode != dir);
2785 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2787 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2793 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2794 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2797 if (!d_is_dir(victim))
2799 if (IS_ROOT(victim))
2801 } else if (d_is_dir(victim))
2803 if (IS_DEADDIR(dir))
2805 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2810 /* Check whether we can create an object with dentry child in directory
2812 * 1. We can't do it if child already exists (open has special treatment for
2813 * this case, but since we are inlined it's OK)
2814 * 2. We can't do it if dir is read-only (done in permission())
2815 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2816 * 4. We should have write and exec permissions on dir
2817 * 5. We can't do it if dir is immutable (done in permission())
2819 static inline int may_create(struct inode *dir, struct dentry *child)
2821 struct user_namespace *s_user_ns;
2822 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2825 if (IS_DEADDIR(dir))
2827 s_user_ns = dir->i_sb->s_user_ns;
2828 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2829 !kgid_has_mapping(s_user_ns, current_fsgid()))
2831 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2835 * p1 and p2 should be directories on the same fs.
2837 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2842 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2846 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2848 p = d_ancestor(p2, p1);
2850 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2851 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2855 p = d_ancestor(p1, p2);
2857 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2858 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2862 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2863 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2866 EXPORT_SYMBOL(lock_rename);
2868 void unlock_rename(struct dentry *p1, struct dentry *p2)
2870 inode_unlock(p1->d_inode);
2872 inode_unlock(p2->d_inode);
2873 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2876 EXPORT_SYMBOL(unlock_rename);
2878 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2881 int error = may_create(dir, dentry);
2885 if (!dir->i_op->create)
2886 return -EACCES; /* shouldn't it be ENOSYS? */
2889 error = security_inode_create(dir, dentry, mode);
2892 error = dir->i_op->create(dir, dentry, mode, want_excl);
2894 fsnotify_create(dir, dentry);
2897 EXPORT_SYMBOL(vfs_create);
2899 bool may_open_dev(const struct path *path)
2901 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2902 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2905 static int may_open(const struct path *path, int acc_mode, int flag)
2907 struct dentry *dentry = path->dentry;
2908 struct inode *inode = dentry->d_inode;
2914 switch (inode->i_mode & S_IFMT) {
2918 if (acc_mode & MAY_WRITE)
2923 if (!may_open_dev(path))
2932 error = inode_permission(inode, MAY_OPEN | acc_mode);
2937 * An append-only file must be opened in append mode for writing.
2939 if (IS_APPEND(inode)) {
2940 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2946 /* O_NOATIME can only be set by the owner or superuser */
2947 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2953 static int handle_truncate(struct file *filp)
2955 const struct path *path = &filp->f_path;
2956 struct inode *inode = path->dentry->d_inode;
2957 int error = get_write_access(inode);
2961 * Refuse to truncate files with mandatory locks held on them.
2963 error = locks_verify_locked(filp);
2965 error = security_path_truncate(path);
2967 error = do_truncate(path->dentry, 0,
2968 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2971 put_write_access(inode);
2975 static inline int open_to_namei_flags(int flag)
2977 if ((flag & O_ACCMODE) == 3)
2982 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2984 struct user_namespace *s_user_ns;
2985 int error = security_path_mknod(dir, dentry, mode, 0);
2989 s_user_ns = dir->dentry->d_sb->s_user_ns;
2990 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2991 !kgid_has_mapping(s_user_ns, current_fsgid()))
2994 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2998 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3002 * Attempt to atomically look up, create and open a file from a negative
3005 * Returns 0 if successful. The file will have been created and attached to
3006 * @file by the filesystem calling finish_open().
3008 * Returns 1 if the file was looked up only or didn't need creating. The
3009 * caller will need to perform the open themselves. @path will have been
3010 * updated to point to the new dentry. This may be negative.
3012 * Returns an error code otherwise.
3014 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3015 struct path *path, struct file *file,
3016 const struct open_flags *op,
3017 int open_flag, umode_t mode,
3020 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3021 struct inode *dir = nd->path.dentry->d_inode;
3024 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3025 open_flag &= ~O_TRUNC;
3027 if (nd->flags & LOOKUP_DIRECTORY)
3028 open_flag |= O_DIRECTORY;
3030 file->f_path.dentry = DENTRY_NOT_SET;
3031 file->f_path.mnt = nd->path.mnt;
3032 error = dir->i_op->atomic_open(dir, dentry, file,
3033 open_to_namei_flags(open_flag),
3035 d_lookup_done(dentry);
3038 * We didn't have the inode before the open, so check open
3041 int acc_mode = op->acc_mode;
3042 if (*opened & FILE_CREATED) {
3043 WARN_ON(!(open_flag & O_CREAT));
3044 fsnotify_create(dir, dentry);
3047 error = may_open(&file->f_path, acc_mode, open_flag);
3048 if (WARN_ON(error > 0))
3050 } else if (error > 0) {
3051 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3054 if (file->f_path.dentry) {
3056 dentry = file->f_path.dentry;
3058 if (*opened & FILE_CREATED)
3059 fsnotify_create(dir, dentry);
3060 if (unlikely(d_is_negative(dentry))) {
3063 path->dentry = dentry;
3064 path->mnt = nd->path.mnt;
3074 * Look up and maybe create and open the last component.
3076 * Must be called with i_mutex held on parent.
3078 * Returns 0 if the file was successfully atomically created (if necessary) and
3079 * opened. In this case the file will be returned attached to @file.
3081 * Returns 1 if the file was not completely opened at this time, though lookups
3082 * and creations will have been performed and the dentry returned in @path will
3083 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3084 * specified then a negative dentry may be returned.
3086 * An error code is returned otherwise.
3088 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3089 * cleared otherwise prior to returning.
3091 static int lookup_open(struct nameidata *nd, struct path *path,
3093 const struct open_flags *op,
3094 bool got_write, int *opened)
3096 struct dentry *dir = nd->path.dentry;
3097 struct inode *dir_inode = dir->d_inode;
3098 int open_flag = op->open_flag;
3099 struct dentry *dentry;
3100 int error, create_error = 0;
3101 umode_t mode = op->mode;
3102 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3104 if (unlikely(IS_DEADDIR(dir_inode)))
3107 *opened &= ~FILE_CREATED;
3108 dentry = d_lookup(dir, &nd->last);
3111 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3113 return PTR_ERR(dentry);
3115 if (d_in_lookup(dentry))
3118 error = d_revalidate(dentry, nd->flags);
3119 if (likely(error > 0))
3123 d_invalidate(dentry);
3127 if (dentry->d_inode) {
3128 /* Cached positive dentry: will open in f_op->open */
3133 * Checking write permission is tricky, bacuse we don't know if we are
3134 * going to actually need it: O_CREAT opens should work as long as the
3135 * file exists. But checking existence breaks atomicity. The trick is
3136 * to check access and if not granted clear O_CREAT from the flags.
3138 * Another problem is returing the "right" error value (e.g. for an
3139 * O_EXCL open we want to return EEXIST not EROFS).
3141 if (open_flag & O_CREAT) {
3142 if (!IS_POSIXACL(dir->d_inode))
3143 mode &= ~current_umask();
3144 if (unlikely(!got_write)) {
3145 create_error = -EROFS;
3146 open_flag &= ~O_CREAT;
3147 if (open_flag & (O_EXCL | O_TRUNC))
3149 /* No side effects, safe to clear O_CREAT */
3151 create_error = may_o_create(&nd->path, dentry, mode);
3153 open_flag &= ~O_CREAT;
3154 if (open_flag & O_EXCL)
3158 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3159 unlikely(!got_write)) {
3161 * No O_CREATE -> atomicity not a requirement -> fall
3162 * back to lookup + open
3167 if (dir_inode->i_op->atomic_open) {
3168 error = atomic_open(nd, dentry, path, file, op, open_flag,
3170 if (unlikely(error == -ENOENT) && create_error)
3171 error = create_error;
3176 if (d_in_lookup(dentry)) {
3177 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3179 d_lookup_done(dentry);
3180 if (unlikely(res)) {
3182 error = PTR_ERR(res);
3190 /* Negative dentry, just create the file */
3191 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3192 *opened |= FILE_CREATED;
3193 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3194 if (!dir_inode->i_op->create) {
3198 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3199 open_flag & O_EXCL);
3202 fsnotify_create(dir_inode, dentry);
3204 if (unlikely(create_error) && !dentry->d_inode) {
3205 error = create_error;
3209 path->dentry = dentry;
3210 path->mnt = nd->path.mnt;
3219 * Handle the last step of open()
3221 static int do_last(struct nameidata *nd,
3222 struct file *file, const struct open_flags *op,
3225 struct dentry *dir = nd->path.dentry;
3226 int open_flag = op->open_flag;
3227 bool will_truncate = (open_flag & O_TRUNC) != 0;
3228 bool got_write = false;
3229 int acc_mode = op->acc_mode;
3231 struct inode *inode;
3235 nd->flags &= ~LOOKUP_PARENT;
3236 nd->flags |= op->intent;
3238 if (nd->last_type != LAST_NORM) {
3239 error = handle_dots(nd, nd->last_type);
3240 if (unlikely(error))
3245 if (!(open_flag & O_CREAT)) {
3246 if (nd->last.name[nd->last.len])
3247 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3248 /* we _can_ be in RCU mode here */
3249 error = lookup_fast(nd, &path, &inode, &seq);
3250 if (likely(error > 0))
3256 BUG_ON(nd->inode != dir->d_inode);
3257 BUG_ON(nd->flags & LOOKUP_RCU);
3259 /* create side of things */
3261 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3262 * has been cleared when we got to the last component we are
3265 error = complete_walk(nd);
3269 audit_inode(nd->name, dir, LOOKUP_PARENT);
3270 /* trailing slashes? */
3271 if (unlikely(nd->last.name[nd->last.len]))
3275 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3276 error = mnt_want_write(nd->path.mnt);
3280 * do _not_ fail yet - we might not need that or fail with
3281 * a different error; let lookup_open() decide; we'll be
3282 * dropping this one anyway.
3285 if (open_flag & O_CREAT)
3286 inode_lock(dir->d_inode);
3288 inode_lock_shared(dir->d_inode);
3289 error = lookup_open(nd, &path, file, op, got_write, opened);
3290 if (open_flag & O_CREAT)
3291 inode_unlock(dir->d_inode);
3293 inode_unlock_shared(dir->d_inode);
3299 if ((*opened & FILE_CREATED) ||
3300 !S_ISREG(file_inode(file)->i_mode))
3301 will_truncate = false;
3303 audit_inode(nd->name, file->f_path.dentry, 0);
3307 if (*opened & FILE_CREATED) {
3308 /* Don't check for write permission, don't truncate */
3309 open_flag &= ~O_TRUNC;
3310 will_truncate = false;
3312 path_to_nameidata(&path, nd);
3313 goto finish_open_created;
3317 * If atomic_open() acquired write access it is dropped now due to
3318 * possible mount and symlink following (this might be optimized away if
3322 mnt_drop_write(nd->path.mnt);
3326 error = follow_managed(&path, nd);
3327 if (unlikely(error < 0))
3330 if (unlikely(d_is_negative(path.dentry))) {
3331 path_to_nameidata(&path, nd);
3336 * create/update audit record if it already exists.
3338 audit_inode(nd->name, path.dentry, 0);
3340 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3341 path_to_nameidata(&path, nd);
3345 seq = 0; /* out of RCU mode, so the value doesn't matter */
3346 inode = d_backing_inode(path.dentry);
3348 error = step_into(nd, &path, 0, inode, seq);
3349 if (unlikely(error))
3352 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3353 error = complete_walk(nd);
3356 audit_inode(nd->name, nd->path.dentry, 0);
3358 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3361 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3363 if (!d_is_reg(nd->path.dentry))
3364 will_truncate = false;
3366 if (will_truncate) {
3367 error = mnt_want_write(nd->path.mnt);
3372 finish_open_created:
3373 error = may_open(&nd->path, acc_mode, open_flag);
3376 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3377 error = vfs_open(&nd->path, file, current_cred());
3380 *opened |= FILE_OPENED;
3382 error = open_check_o_direct(file);
3384 error = ima_file_check(file, op->acc_mode, *opened);
3385 if (!error && will_truncate)
3386 error = handle_truncate(file);
3388 if (unlikely(error) && (*opened & FILE_OPENED))
3390 if (unlikely(error > 0)) {
3395 mnt_drop_write(nd->path.mnt);
3399 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3401 struct dentry *child = NULL;
3402 struct inode *dir = dentry->d_inode;
3403 struct inode *inode;
3406 /* we want directory to be writable */
3407 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3410 error = -EOPNOTSUPP;
3411 if (!dir->i_op->tmpfile)
3414 child = d_alloc(dentry, &slash_name);
3415 if (unlikely(!child))
3417 error = dir->i_op->tmpfile(dir, child, mode);
3421 inode = child->d_inode;
3422 if (unlikely(!inode))
3424 if (!(open_flag & O_EXCL)) {
3425 spin_lock(&inode->i_lock);
3426 inode->i_state |= I_LINKABLE;
3427 spin_unlock(&inode->i_lock);
3433 return ERR_PTR(error);
3435 EXPORT_SYMBOL(vfs_tmpfile);
3437 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3438 const struct open_flags *op,
3439 struct file *file, int *opened)
3441 struct dentry *child;
3443 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3444 if (unlikely(error))
3446 error = mnt_want_write(path.mnt);
3447 if (unlikely(error))
3449 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3450 error = PTR_ERR(child);
3451 if (unlikely(IS_ERR(child)))
3454 path.dentry = child;
3455 audit_inode(nd->name, child, 0);
3456 /* Don't check for other permissions, the inode was just created */
3457 error = may_open(&path, 0, op->open_flag);
3460 file->f_path.mnt = path.mnt;
3461 error = finish_open(file, child, NULL, opened);
3464 error = open_check_o_direct(file);
3468 mnt_drop_write(path.mnt);
3474 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3477 int error = path_lookupat(nd, flags, &path);
3479 audit_inode(nd->name, path.dentry, 0);
3480 error = vfs_open(&path, file, current_cred());
3486 static struct file *path_openat(struct nameidata *nd,
3487 const struct open_flags *op, unsigned flags)
3494 file = get_empty_filp();
3498 file->f_flags = op->open_flag;
3500 if (unlikely(file->f_flags & __O_TMPFILE)) {
3501 error = do_tmpfile(nd, flags, op, file, &opened);
3505 if (unlikely(file->f_flags & O_PATH)) {
3506 error = do_o_path(nd, flags, file);
3508 opened |= FILE_OPENED;
3512 s = path_init(nd, flags);
3517 while (!(error = link_path_walk(s, nd)) &&
3518 (error = do_last(nd, file, op, &opened)) > 0) {
3519 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3520 s = trailing_symlink(nd);
3528 if (!(opened & FILE_OPENED)) {
3532 if (unlikely(error)) {
3533 if (error == -EOPENSTALE) {
3534 if (flags & LOOKUP_RCU)
3539 file = ERR_PTR(error);
3544 struct file *do_filp_open(int dfd, struct filename *pathname,
3545 const struct open_flags *op)
3547 struct nameidata nd;
3548 int flags = op->lookup_flags;
3551 set_nameidata(&nd, dfd, pathname);
3552 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3553 if (unlikely(filp == ERR_PTR(-ECHILD)))
3554 filp = path_openat(&nd, op, flags);
3555 if (unlikely(filp == ERR_PTR(-ESTALE)))
3556 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3557 restore_nameidata();
3561 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3562 const char *name, const struct open_flags *op)
3564 struct nameidata nd;
3566 struct filename *filename;
3567 int flags = op->lookup_flags | LOOKUP_ROOT;
3570 nd.root.dentry = dentry;
3572 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3573 return ERR_PTR(-ELOOP);
3575 filename = getname_kernel(name);
3576 if (IS_ERR(filename))
3577 return ERR_CAST(filename);
3579 set_nameidata(&nd, -1, filename);
3580 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3581 if (unlikely(file == ERR_PTR(-ECHILD)))
3582 file = path_openat(&nd, op, flags);
3583 if (unlikely(file == ERR_PTR(-ESTALE)))
3584 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3585 restore_nameidata();
3590 static struct dentry *filename_create(int dfd, struct filename *name,
3591 struct path *path, unsigned int lookup_flags)
3593 struct dentry *dentry = ERR_PTR(-EEXIST);
3598 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3601 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3602 * other flags passed in are ignored!
3604 lookup_flags &= LOOKUP_REVAL;
3606 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3608 return ERR_CAST(name);
3611 * Yucky last component or no last component at all?
3612 * (foo/., foo/.., /////)
3614 if (unlikely(type != LAST_NORM))
3617 /* don't fail immediately if it's r/o, at least try to report other errors */
3618 err2 = mnt_want_write(path->mnt);
3620 * Do the final lookup.
3622 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3623 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3624 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3629 if (d_is_positive(dentry))
3633 * Special case - lookup gave negative, but... we had foo/bar/
3634 * From the vfs_mknod() POV we just have a negative dentry -
3635 * all is fine. Let's be bastards - you had / on the end, you've
3636 * been asking for (non-existent) directory. -ENOENT for you.
3638 if (unlikely(!is_dir && last.name[last.len])) {
3642 if (unlikely(err2)) {
3650 dentry = ERR_PTR(error);
3652 inode_unlock(path->dentry->d_inode);
3654 mnt_drop_write(path->mnt);
3661 struct dentry *kern_path_create(int dfd, const char *pathname,
3662 struct path *path, unsigned int lookup_flags)
3664 return filename_create(dfd, getname_kernel(pathname),
3665 path, lookup_flags);
3667 EXPORT_SYMBOL(kern_path_create);
3669 void done_path_create(struct path *path, struct dentry *dentry)
3672 inode_unlock(path->dentry->d_inode);
3673 mnt_drop_write(path->mnt);
3676 EXPORT_SYMBOL(done_path_create);
3678 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3679 struct path *path, unsigned int lookup_flags)
3681 return filename_create(dfd, getname(pathname), path, lookup_flags);
3683 EXPORT_SYMBOL(user_path_create);
3685 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3687 int error = may_create(dir, dentry);
3692 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3695 if (!dir->i_op->mknod)
3698 error = devcgroup_inode_mknod(mode, dev);
3702 error = security_inode_mknod(dir, dentry, mode, dev);
3706 error = dir->i_op->mknod(dir, dentry, mode, dev);
3708 fsnotify_create(dir, dentry);
3711 EXPORT_SYMBOL(vfs_mknod);
3713 static int may_mknod(umode_t mode)
3715 switch (mode & S_IFMT) {
3721 case 0: /* zero mode translates to S_IFREG */
3730 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3733 struct dentry *dentry;
3736 unsigned int lookup_flags = 0;
3738 error = may_mknod(mode);
3742 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3744 return PTR_ERR(dentry);
3746 if (!IS_POSIXACL(path.dentry->d_inode))
3747 mode &= ~current_umask();
3748 error = security_path_mknod(&path, dentry, mode, dev);
3751 switch (mode & S_IFMT) {
3752 case 0: case S_IFREG:
3753 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3755 ima_post_path_mknod(dentry);
3757 case S_IFCHR: case S_IFBLK:
3758 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3759 new_decode_dev(dev));
3761 case S_IFIFO: case S_IFSOCK:
3762 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3766 done_path_create(&path, dentry);
3767 if (retry_estale(error, lookup_flags)) {
3768 lookup_flags |= LOOKUP_REVAL;
3774 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3776 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3779 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3781 int error = may_create(dir, dentry);
3782 unsigned max_links = dir->i_sb->s_max_links;
3787 if (!dir->i_op->mkdir)
3790 mode &= (S_IRWXUGO|S_ISVTX);
3791 error = security_inode_mkdir(dir, dentry, mode);
3795 if (max_links && dir->i_nlink >= max_links)
3798 error = dir->i_op->mkdir(dir, dentry, mode);
3800 fsnotify_mkdir(dir, dentry);
3803 EXPORT_SYMBOL(vfs_mkdir);
3805 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3807 struct dentry *dentry;
3810 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3813 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3815 return PTR_ERR(dentry);
3817 if (!IS_POSIXACL(path.dentry->d_inode))
3818 mode &= ~current_umask();
3819 error = security_path_mkdir(&path, dentry, mode);
3821 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3822 done_path_create(&path, dentry);
3823 if (retry_estale(error, lookup_flags)) {
3824 lookup_flags |= LOOKUP_REVAL;
3830 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3832 return sys_mkdirat(AT_FDCWD, pathname, mode);
3835 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3837 int error = may_delete(dir, dentry, 1);
3842 if (!dir->i_op->rmdir)
3846 inode_lock(dentry->d_inode);
3849 if (is_local_mountpoint(dentry))
3852 error = security_inode_rmdir(dir, dentry);
3856 shrink_dcache_parent(dentry);
3857 error = dir->i_op->rmdir(dir, dentry);
3861 dentry->d_inode->i_flags |= S_DEAD;
3863 detach_mounts(dentry);
3866 inode_unlock(dentry->d_inode);
3872 EXPORT_SYMBOL(vfs_rmdir);
3874 static long do_rmdir(int dfd, const char __user *pathname)
3877 struct filename *name;
3878 struct dentry *dentry;
3882 unsigned int lookup_flags = 0;
3884 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3885 &path, &last, &type);
3887 return PTR_ERR(name);
3901 error = mnt_want_write(path.mnt);
3905 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3906 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3907 error = PTR_ERR(dentry);
3910 if (!dentry->d_inode) {
3914 error = security_path_rmdir(&path, dentry);
3917 error = vfs_rmdir(path.dentry->d_inode, dentry);
3921 inode_unlock(path.dentry->d_inode);
3922 mnt_drop_write(path.mnt);
3926 if (retry_estale(error, lookup_flags)) {
3927 lookup_flags |= LOOKUP_REVAL;
3933 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3935 return do_rmdir(AT_FDCWD, pathname);
3939 * vfs_unlink - unlink a filesystem object
3940 * @dir: parent directory
3942 * @delegated_inode: returns victim inode, if the inode is delegated.
3944 * The caller must hold dir->i_mutex.
3946 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3947 * return a reference to the inode in delegated_inode. The caller
3948 * should then break the delegation on that inode and retry. Because
3949 * breaking a delegation may take a long time, the caller should drop
3950 * dir->i_mutex before doing so.
3952 * Alternatively, a caller may pass NULL for delegated_inode. This may
3953 * be appropriate for callers that expect the underlying filesystem not
3954 * to be NFS exported.
3956 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3958 struct inode *target = dentry->d_inode;
3959 int error = may_delete(dir, dentry, 0);
3964 if (!dir->i_op->unlink)
3968 if (is_local_mountpoint(dentry))
3971 error = security_inode_unlink(dir, dentry);
3973 error = try_break_deleg(target, delegated_inode);
3976 error = dir->i_op->unlink(dir, dentry);
3979 detach_mounts(dentry);
3984 inode_unlock(target);
3986 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3987 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3988 fsnotify_link_count(target);
3994 EXPORT_SYMBOL(vfs_unlink);
3997 * Make sure that the actual truncation of the file will occur outside its
3998 * directory's i_mutex. Truncate can take a long time if there is a lot of
3999 * writeout happening, and we don't want to prevent access to the directory
4000 * while waiting on the I/O.
4002 static long do_unlinkat(int dfd, const char __user *pathname)
4005 struct filename *name;
4006 struct dentry *dentry;
4010 struct inode *inode = NULL;
4011 struct inode *delegated_inode = NULL;
4012 unsigned int lookup_flags = 0;
4014 name = filename_parentat(dfd, getname(pathname), lookup_flags,
4015 &path, &last, &type);
4017 return PTR_ERR(name);
4020 if (type != LAST_NORM)
4023 error = mnt_want_write(path.mnt);
4027 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4028 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4029 error = PTR_ERR(dentry);
4030 if (!IS_ERR(dentry)) {
4031 /* Why not before? Because we want correct error value */
4032 if (last.name[last.len])
4034 inode = dentry->d_inode;
4035 if (d_is_negative(dentry))
4038 error = security_path_unlink(&path, dentry);
4041 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4045 inode_unlock(path.dentry->d_inode);
4047 iput(inode); /* truncate the inode here */
4049 if (delegated_inode) {
4050 error = break_deleg_wait(&delegated_inode);
4054 mnt_drop_write(path.mnt);
4058 if (retry_estale(error, lookup_flags)) {
4059 lookup_flags |= LOOKUP_REVAL;
4066 if (d_is_negative(dentry))
4068 else if (d_is_dir(dentry))
4075 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4077 if ((flag & ~AT_REMOVEDIR) != 0)
4080 if (flag & AT_REMOVEDIR)
4081 return do_rmdir(dfd, pathname);
4083 return do_unlinkat(dfd, pathname);
4086 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4088 return do_unlinkat(AT_FDCWD, pathname);
4091 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4093 int error = may_create(dir, dentry);
4098 if (!dir->i_op->symlink)
4101 error = security_inode_symlink(dir, dentry, oldname);
4105 error = dir->i_op->symlink(dir, dentry, oldname);
4107 fsnotify_create(dir, dentry);
4110 EXPORT_SYMBOL(vfs_symlink);
4112 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4113 int, newdfd, const char __user *, newname)
4116 struct filename *from;
4117 struct dentry *dentry;
4119 unsigned int lookup_flags = 0;
4121 from = getname(oldname);
4123 return PTR_ERR(from);
4125 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4126 error = PTR_ERR(dentry);
4130 error = security_path_symlink(&path, dentry, from->name);
4132 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4133 done_path_create(&path, dentry);
4134 if (retry_estale(error, lookup_flags)) {
4135 lookup_flags |= LOOKUP_REVAL;
4143 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4145 return sys_symlinkat(oldname, AT_FDCWD, newname);
4149 * vfs_link - create a new link
4150 * @old_dentry: object to be linked
4152 * @new_dentry: where to create the new link
4153 * @delegated_inode: returns inode needing a delegation break
4155 * The caller must hold dir->i_mutex
4157 * If vfs_link discovers a delegation on the to-be-linked file in need
4158 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4159 * inode in delegated_inode. The caller should then break the delegation
4160 * and retry. Because breaking a delegation may take a long time, the
4161 * caller should drop the i_mutex before doing so.
4163 * Alternatively, a caller may pass NULL for delegated_inode. This may
4164 * be appropriate for callers that expect the underlying filesystem not
4165 * to be NFS exported.
4167 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4169 struct inode *inode = old_dentry->d_inode;
4170 unsigned max_links = dir->i_sb->s_max_links;
4176 error = may_create(dir, new_dentry);
4180 if (dir->i_sb != inode->i_sb)
4184 * A link to an append-only or immutable file cannot be created.
4186 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4189 * Updating the link count will likely cause i_uid and i_gid to
4190 * be writen back improperly if their true value is unknown to
4193 if (HAS_UNMAPPED_ID(inode))
4195 if (!dir->i_op->link)
4197 if (S_ISDIR(inode->i_mode))
4200 error = security_inode_link(old_dentry, dir, new_dentry);
4205 /* Make sure we don't allow creating hardlink to an unlinked file */
4206 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4208 else if (max_links && inode->i_nlink >= max_links)
4211 error = try_break_deleg(inode, delegated_inode);
4213 error = dir->i_op->link(old_dentry, dir, new_dentry);
4216 if (!error && (inode->i_state & I_LINKABLE)) {
4217 spin_lock(&inode->i_lock);
4218 inode->i_state &= ~I_LINKABLE;
4219 spin_unlock(&inode->i_lock);
4221 inode_unlock(inode);
4223 fsnotify_link(dir, inode, new_dentry);
4226 EXPORT_SYMBOL(vfs_link);
4229 * Hardlinks are often used in delicate situations. We avoid
4230 * security-related surprises by not following symlinks on the
4233 * We don't follow them on the oldname either to be compatible
4234 * with linux 2.0, and to avoid hard-linking to directories
4235 * and other special files. --ADM
4237 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4238 int, newdfd, const char __user *, newname, int, flags)
4240 struct dentry *new_dentry;
4241 struct path old_path, new_path;
4242 struct inode *delegated_inode = NULL;
4246 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4249 * To use null names we require CAP_DAC_READ_SEARCH
4250 * This ensures that not everyone will be able to create
4251 * handlink using the passed filedescriptor.
4253 if (flags & AT_EMPTY_PATH) {
4254 if (!capable(CAP_DAC_READ_SEARCH))
4259 if (flags & AT_SYMLINK_FOLLOW)
4260 how |= LOOKUP_FOLLOW;
4262 error = user_path_at(olddfd, oldname, how, &old_path);
4266 new_dentry = user_path_create(newdfd, newname, &new_path,
4267 (how & LOOKUP_REVAL));
4268 error = PTR_ERR(new_dentry);
4269 if (IS_ERR(new_dentry))
4273 if (old_path.mnt != new_path.mnt)
4275 error = may_linkat(&old_path);
4276 if (unlikely(error))
4278 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4281 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4283 done_path_create(&new_path, new_dentry);
4284 if (delegated_inode) {
4285 error = break_deleg_wait(&delegated_inode);
4287 path_put(&old_path);
4291 if (retry_estale(error, how)) {
4292 path_put(&old_path);
4293 how |= LOOKUP_REVAL;
4297 path_put(&old_path);
4302 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4304 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4308 * vfs_rename - rename a filesystem object
4309 * @old_dir: parent of source
4310 * @old_dentry: source
4311 * @new_dir: parent of destination
4312 * @new_dentry: destination
4313 * @delegated_inode: returns an inode needing a delegation break
4314 * @flags: rename flags
4316 * The caller must hold multiple mutexes--see lock_rename()).
4318 * If vfs_rename discovers a delegation in need of breaking at either
4319 * the source or destination, it will return -EWOULDBLOCK and return a
4320 * reference to the inode in delegated_inode. The caller should then
4321 * break the delegation and retry. Because breaking a delegation may
4322 * take a long time, the caller should drop all locks before doing
4325 * Alternatively, a caller may pass NULL for delegated_inode. This may
4326 * be appropriate for callers that expect the underlying filesystem not
4327 * to be NFS exported.
4329 * The worst of all namespace operations - renaming directory. "Perverted"
4330 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4333 * a) we can get into loop creation.
4334 * b) race potential - two innocent renames can create a loop together.
4335 * That's where 4.4 screws up. Current fix: serialization on
4336 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4338 * c) we have to lock _four_ objects - parents and victim (if it exists),
4339 * and source (if it is not a directory).
4340 * And that - after we got ->i_mutex on parents (until then we don't know
4341 * whether the target exists). Solution: try to be smart with locking
4342 * order for inodes. We rely on the fact that tree topology may change
4343 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4344 * move will be locked. Thus we can rank directories by the tree
4345 * (ancestors first) and rank all non-directories after them.
4346 * That works since everybody except rename does "lock parent, lookup,
4347 * lock child" and rename is under ->s_vfs_rename_mutex.
4348 * HOWEVER, it relies on the assumption that any object with ->lookup()
4349 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4350 * we'd better make sure that there's no link(2) for them.
4351 * d) conversion from fhandle to dentry may come in the wrong moment - when
4352 * we are removing the target. Solution: we will have to grab ->i_mutex
4353 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4354 * ->i_mutex on parents, which works but leads to some truly excessive
4357 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4358 struct inode *new_dir, struct dentry *new_dentry,
4359 struct inode **delegated_inode, unsigned int flags)
4362 bool is_dir = d_is_dir(old_dentry);
4363 struct inode *source = old_dentry->d_inode;
4364 struct inode *target = new_dentry->d_inode;
4365 bool new_is_dir = false;
4366 unsigned max_links = new_dir->i_sb->s_max_links;
4367 struct name_snapshot old_name;
4369 if (source == target)
4372 error = may_delete(old_dir, old_dentry, is_dir);
4377 error = may_create(new_dir, new_dentry);
4379 new_is_dir = d_is_dir(new_dentry);
4381 if (!(flags & RENAME_EXCHANGE))
4382 error = may_delete(new_dir, new_dentry, is_dir);
4384 error = may_delete(new_dir, new_dentry, new_is_dir);
4389 if (!old_dir->i_op->rename)
4393 * If we are going to change the parent - check write permissions,
4394 * we'll need to flip '..'.
4396 if (new_dir != old_dir) {
4398 error = inode_permission(source, MAY_WRITE);
4402 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4403 error = inode_permission(target, MAY_WRITE);
4409 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4414 take_dentry_name_snapshot(&old_name, old_dentry);
4416 if (!is_dir || (flags & RENAME_EXCHANGE))
4417 lock_two_nondirectories(source, target);
4422 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4425 if (max_links && new_dir != old_dir) {
4427 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4429 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4430 old_dir->i_nlink >= max_links)
4433 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4434 shrink_dcache_parent(new_dentry);
4436 error = try_break_deleg(source, delegated_inode);
4440 if (target && !new_is_dir) {
4441 error = try_break_deleg(target, delegated_inode);
4445 error = old_dir->i_op->rename(old_dir, old_dentry,
4446 new_dir, new_dentry, flags);
4450 if (!(flags & RENAME_EXCHANGE) && target) {
4452 target->i_flags |= S_DEAD;
4453 dont_mount(new_dentry);
4454 detach_mounts(new_dentry);
4456 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4457 if (!(flags & RENAME_EXCHANGE))
4458 d_move(old_dentry, new_dentry);
4460 d_exchange(old_dentry, new_dentry);
4463 if (!is_dir || (flags & RENAME_EXCHANGE))
4464 unlock_two_nondirectories(source, target);
4466 inode_unlock(target);
4469 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4470 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4471 if (flags & RENAME_EXCHANGE) {
4472 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4473 new_is_dir, NULL, new_dentry);
4476 release_dentry_name_snapshot(&old_name);
4480 EXPORT_SYMBOL(vfs_rename);
4482 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4483 int, newdfd, const char __user *, newname, unsigned int, flags)
4485 struct dentry *old_dentry, *new_dentry;
4486 struct dentry *trap;
4487 struct path old_path, new_path;
4488 struct qstr old_last, new_last;
4489 int old_type, new_type;
4490 struct inode *delegated_inode = NULL;
4491 struct filename *from;
4492 struct filename *to;
4493 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4494 bool should_retry = false;
4497 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4500 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4501 (flags & RENAME_EXCHANGE))
4504 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4507 if (flags & RENAME_EXCHANGE)
4511 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4512 &old_path, &old_last, &old_type);
4514 error = PTR_ERR(from);
4518 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4519 &new_path, &new_last, &new_type);
4521 error = PTR_ERR(to);
4526 if (old_path.mnt != new_path.mnt)
4530 if (old_type != LAST_NORM)
4533 if (flags & RENAME_NOREPLACE)
4535 if (new_type != LAST_NORM)
4538 error = mnt_want_write(old_path.mnt);
4543 trap = lock_rename(new_path.dentry, old_path.dentry);
4545 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4546 error = PTR_ERR(old_dentry);
4547 if (IS_ERR(old_dentry))
4549 /* source must exist */
4551 if (d_is_negative(old_dentry))
4553 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4554 error = PTR_ERR(new_dentry);
4555 if (IS_ERR(new_dentry))
4558 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4560 if (flags & RENAME_EXCHANGE) {
4562 if (d_is_negative(new_dentry))
4565 if (!d_is_dir(new_dentry)) {
4567 if (new_last.name[new_last.len])
4571 /* unless the source is a directory trailing slashes give -ENOTDIR */
4572 if (!d_is_dir(old_dentry)) {
4574 if (old_last.name[old_last.len])
4576 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4579 /* source should not be ancestor of target */
4581 if (old_dentry == trap)
4583 /* target should not be an ancestor of source */
4584 if (!(flags & RENAME_EXCHANGE))
4586 if (new_dentry == trap)
4589 error = security_path_rename(&old_path, old_dentry,
4590 &new_path, new_dentry, flags);
4593 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4594 new_path.dentry->d_inode, new_dentry,
4595 &delegated_inode, flags);
4601 unlock_rename(new_path.dentry, old_path.dentry);
4602 if (delegated_inode) {
4603 error = break_deleg_wait(&delegated_inode);
4607 mnt_drop_write(old_path.mnt);
4609 if (retry_estale(error, lookup_flags))
4610 should_retry = true;
4611 path_put(&new_path);
4614 path_put(&old_path);
4617 should_retry = false;
4618 lookup_flags |= LOOKUP_REVAL;
4625 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4626 int, newdfd, const char __user *, newname)
4628 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4631 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4633 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4636 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4638 int error = may_create(dir, dentry);
4642 if (!dir->i_op->mknod)
4645 return dir->i_op->mknod(dir, dentry,
4646 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4648 EXPORT_SYMBOL(vfs_whiteout);
4650 int readlink_copy(char __user *buffer, int buflen, const char *link)
4652 int len = PTR_ERR(link);
4657 if (len > (unsigned) buflen)
4659 if (copy_to_user(buffer, link, len))
4666 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4667 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4668 * for any given inode is up to filesystem.
4670 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4673 DEFINE_DELAYED_CALL(done);
4674 struct inode *inode = d_inode(dentry);
4675 const char *link = inode->i_link;
4679 link = inode->i_op->get_link(dentry, inode, &done);
4681 return PTR_ERR(link);
4683 res = readlink_copy(buffer, buflen, link);
4684 do_delayed_call(&done);
4689 * vfs_readlink - copy symlink body into userspace buffer
4690 * @dentry: dentry on which to get symbolic link
4691 * @buffer: user memory pointer
4692 * @buflen: size of buffer
4694 * Does not touch atime. That's up to the caller if necessary
4696 * Does not call security hook.
4698 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4700 struct inode *inode = d_inode(dentry);
4702 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4703 if (unlikely(inode->i_op->readlink))
4704 return inode->i_op->readlink(dentry, buffer, buflen);
4706 if (!d_is_symlink(dentry))
4709 spin_lock(&inode->i_lock);
4710 inode->i_opflags |= IOP_DEFAULT_READLINK;
4711 spin_unlock(&inode->i_lock);
4714 return generic_readlink(dentry, buffer, buflen);
4716 EXPORT_SYMBOL(vfs_readlink);
4719 * vfs_get_link - get symlink body
4720 * @dentry: dentry on which to get symbolic link
4721 * @done: caller needs to free returned data with this
4723 * Calls security hook and i_op->get_link() on the supplied inode.
4725 * It does not touch atime. That's up to the caller if necessary.
4727 * Does not work on "special" symlinks like /proc/$$/fd/N
4729 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4731 const char *res = ERR_PTR(-EINVAL);
4732 struct inode *inode = d_inode(dentry);
4734 if (d_is_symlink(dentry)) {
4735 res = ERR_PTR(security_inode_readlink(dentry));
4737 res = inode->i_op->get_link(dentry, inode, done);
4741 EXPORT_SYMBOL(vfs_get_link);
4743 /* get the link contents into pagecache */
4744 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4745 struct delayed_call *callback)
4749 struct address_space *mapping = inode->i_mapping;
4752 page = find_get_page(mapping, 0);
4754 return ERR_PTR(-ECHILD);
4755 if (!PageUptodate(page)) {
4757 return ERR_PTR(-ECHILD);
4760 page = read_mapping_page(mapping, 0, NULL);
4764 set_delayed_call(callback, page_put_link, page);
4765 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4766 kaddr = page_address(page);
4767 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4771 EXPORT_SYMBOL(page_get_link);
4773 void page_put_link(void *arg)
4777 EXPORT_SYMBOL(page_put_link);
4779 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4781 DEFINE_DELAYED_CALL(done);
4782 int res = readlink_copy(buffer, buflen,
4783 page_get_link(dentry, d_inode(dentry),
4785 do_delayed_call(&done);
4788 EXPORT_SYMBOL(page_readlink);
4791 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4793 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4795 struct address_space *mapping = inode->i_mapping;
4799 unsigned int flags = 0;
4801 flags |= AOP_FLAG_NOFS;
4804 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4805 flags, &page, &fsdata);
4809 memcpy(page_address(page), symname, len-1);
4811 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4818 mark_inode_dirty(inode);
4823 EXPORT_SYMBOL(__page_symlink);
4825 int page_symlink(struct inode *inode, const char *symname, int len)
4827 return __page_symlink(inode, symname, len,
4828 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4830 EXPORT_SYMBOL(page_symlink);
4832 const struct inode_operations page_symlink_inode_operations = {
4833 .get_link = page_get_link,
4835 EXPORT_SYMBOL(page_symlink_inode_operations);