1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
262 static int check_acl(struct inode *inode, int mask)
264 #ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
274 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
291 * This does the basic permission checking
293 static int acl_permission_check(struct inode *inode, int mask)
295 unsigned int mode = inode->i_mode;
297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
301 int error = check_acl(inode, mask);
302 if (error != -EAGAIN)
306 if (in_group_p(inode->i_gid))
311 * If the DACs are ok we don't need any capability check.
313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
332 int generic_permission(struct inode *inode, int mask)
337 * Do the basic permission checks.
339 ret = acl_permission_check(inode, mask);
343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
355 * Searching includes executable on directories, else just read.
357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
358 if (mask == MAY_READ)
359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
372 EXPORT_SYMBOL(generic_permission);
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
380 static inline int do_inode_permission(struct inode *inode, int mask)
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
391 return generic_permission(inode, mask);
395 * __inode_permission - Check for access rights to a given inode
396 * @inode: Inode to check permission on
397 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
399 * Check for read/write/execute permissions on an inode.
401 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
403 * This does not check for a read-only file system. You probably want
404 * inode_permission().
406 int __inode_permission(struct inode *inode, int mask)
410 if (unlikely(mask & MAY_WRITE)) {
412 * Nobody gets write access to an immutable file.
414 if (IS_IMMUTABLE(inode))
418 * Updating mtime will likely cause i_uid and i_gid to be
419 * written back improperly if their true value is unknown
422 if (HAS_UNMAPPED_ID(inode))
426 retval = do_inode_permission(inode, mask);
430 retval = devcgroup_inode_permission(inode, mask);
434 return security_inode_permission(inode, mask);
436 EXPORT_SYMBOL(__inode_permission);
439 * sb_permission - Check superblock-level permissions
440 * @sb: Superblock of inode to check permission on
441 * @inode: Inode to check permission on
442 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
444 * Separate out file-system wide checks from inode-specific permission checks.
446 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
448 if (unlikely(mask & MAY_WRITE)) {
449 umode_t mode = inode->i_mode;
451 /* Nobody gets write access to a read-only fs. */
452 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
459 * inode_permission - Check for access rights to a given inode
460 * @inode: Inode to check permission on
461 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
463 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
464 * this, letting us set arbitrary permissions for filesystem access without
465 * changing the "normal" UIDs which are used for other things.
467 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
469 int inode_permission(struct inode *inode, int mask)
473 retval = sb_permission(inode->i_sb, inode, mask);
476 return __inode_permission(inode, mask);
478 EXPORT_SYMBOL(inode_permission);
481 * path_get - get a reference to a path
482 * @path: path to get the reference to
484 * Given a path increment the reference count to the dentry and the vfsmount.
486 void path_get(const struct path *path)
491 EXPORT_SYMBOL(path_get);
494 * path_put - put a reference to a path
495 * @path: path to put the reference to
497 * Given a path decrement the reference count to the dentry and the vfsmount.
499 void path_put(const struct path *path)
504 EXPORT_SYMBOL(path_put);
506 #define EMBEDDED_LEVELS 2
511 struct inode *inode; /* path.dentry.d_inode */
516 int total_link_count;
519 struct delayed_call done;
522 } *stack, internal[EMBEDDED_LEVELS];
523 struct filename *name;
524 struct nameidata *saved;
525 struct inode *link_inode;
528 } __randomize_layout;
530 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
532 struct nameidata *old = current->nameidata;
533 p->stack = p->internal;
536 p->total_link_count = old ? old->total_link_count : 0;
538 current->nameidata = p;
541 static void restore_nameidata(void)
543 struct nameidata *now = current->nameidata, *old = now->saved;
545 current->nameidata = old;
547 old->total_link_count = now->total_link_count;
548 if (now->stack != now->internal)
552 static int __nd_alloc_stack(struct nameidata *nd)
556 if (nd->flags & LOOKUP_RCU) {
557 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
562 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
567 memcpy(p, nd->internal, sizeof(nd->internal));
573 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
574 * @path: nameidate to verify
576 * Rename can sometimes move a file or directory outside of a bind
577 * mount, path_connected allows those cases to be detected.
579 static bool path_connected(const struct path *path)
581 struct vfsmount *mnt = path->mnt;
582 struct super_block *sb = mnt->mnt_sb;
584 /* Bind mounts and multi-root filesystems can have disconnected paths */
585 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
588 return is_subdir(path->dentry, mnt->mnt_root);
591 static inline int nd_alloc_stack(struct nameidata *nd)
593 if (likely(nd->depth != EMBEDDED_LEVELS))
595 if (likely(nd->stack != nd->internal))
597 return __nd_alloc_stack(nd);
600 static void drop_links(struct nameidata *nd)
604 struct saved *last = nd->stack + i;
605 do_delayed_call(&last->done);
606 clear_delayed_call(&last->done);
610 static void terminate_walk(struct nameidata *nd)
613 if (!(nd->flags & LOOKUP_RCU)) {
616 for (i = 0; i < nd->depth; i++)
617 path_put(&nd->stack[i].link);
618 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
623 nd->flags &= ~LOOKUP_RCU;
624 if (!(nd->flags & LOOKUP_ROOT))
631 /* path_put is needed afterwards regardless of success or failure */
632 static bool legitimize_path(struct nameidata *nd,
633 struct path *path, unsigned seq)
635 int res = __legitimize_mnt(path->mnt, nd->m_seq);
642 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
646 return !read_seqcount_retry(&path->dentry->d_seq, seq);
649 static bool legitimize_links(struct nameidata *nd)
652 for (i = 0; i < nd->depth; i++) {
653 struct saved *last = nd->stack + i;
654 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
664 * Path walking has 2 modes, rcu-walk and ref-walk (see
665 * Documentation/filesystems/path-lookup.txt). In situations when we can't
666 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
667 * normal reference counts on dentries and vfsmounts to transition to ref-walk
668 * mode. Refcounts are grabbed at the last known good point before rcu-walk
669 * got stuck, so ref-walk may continue from there. If this is not successful
670 * (eg. a seqcount has changed), then failure is returned and it's up to caller
671 * to restart the path walk from the beginning in ref-walk mode.
675 * unlazy_walk - try to switch to ref-walk mode.
676 * @nd: nameidata pathwalk data
677 * Returns: 0 on success, -ECHILD on failure
679 * unlazy_walk attempts to legitimize the current nd->path and nd->root
681 * Must be called from rcu-walk context.
682 * Nothing should touch nameidata between unlazy_walk() failure and
685 static int unlazy_walk(struct nameidata *nd)
687 struct dentry *parent = nd->path.dentry;
689 BUG_ON(!(nd->flags & LOOKUP_RCU));
691 nd->flags &= ~LOOKUP_RCU;
692 if (unlikely(!legitimize_links(nd)))
694 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
696 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
697 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
701 BUG_ON(nd->inode != parent->d_inode);
706 nd->path.dentry = NULL;
708 if (!(nd->flags & LOOKUP_ROOT))
716 * unlazy_child - try to switch to ref-walk mode.
717 * @nd: nameidata pathwalk data
718 * @dentry: child of nd->path.dentry
719 * @seq: seq number to check dentry against
720 * Returns: 0 on success, -ECHILD on failure
722 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
723 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
724 * @nd. Must be called from rcu-walk context.
725 * Nothing should touch nameidata between unlazy_child() failure and
728 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
730 BUG_ON(!(nd->flags & LOOKUP_RCU));
732 nd->flags &= ~LOOKUP_RCU;
733 if (unlikely(!legitimize_links(nd)))
735 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
737 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
741 * We need to move both the parent and the dentry from the RCU domain
742 * to be properly refcounted. And the sequence number in the dentry
743 * validates *both* dentry counters, since we checked the sequence
744 * number of the parent after we got the child sequence number. So we
745 * know the parent must still be valid if the child sequence number is
747 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
749 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
755 * Sequence counts matched. Now make sure that the root is
756 * still valid and get it if required.
758 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
759 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
772 nd->path.dentry = NULL;
776 if (!(nd->flags & LOOKUP_ROOT))
781 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
783 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
784 return dentry->d_op->d_revalidate(dentry, flags);
790 * complete_walk - successful completion of path walk
791 * @nd: pointer nameidata
793 * If we had been in RCU mode, drop out of it and legitimize nd->path.
794 * Revalidate the final result, unless we'd already done that during
795 * the path walk or the filesystem doesn't ask for it. Return 0 on
796 * success, -error on failure. In case of failure caller does not
797 * need to drop nd->path.
799 static int complete_walk(struct nameidata *nd)
801 struct dentry *dentry = nd->path.dentry;
804 if (nd->flags & LOOKUP_RCU) {
805 if (!(nd->flags & LOOKUP_ROOT))
807 if (unlikely(unlazy_walk(nd)))
811 if (likely(!(nd->flags & LOOKUP_JUMPED)))
814 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
817 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
827 static void set_root(struct nameidata *nd)
829 struct fs_struct *fs = current->fs;
831 if (nd->flags & LOOKUP_RCU) {
835 seq = read_seqcount_begin(&fs->seq);
837 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
838 } while (read_seqcount_retry(&fs->seq, seq));
840 get_fs_root(fs, &nd->root);
844 static void path_put_conditional(struct path *path, struct nameidata *nd)
847 if (path->mnt != nd->path.mnt)
851 static inline void path_to_nameidata(const struct path *path,
852 struct nameidata *nd)
854 if (!(nd->flags & LOOKUP_RCU)) {
855 dput(nd->path.dentry);
856 if (nd->path.mnt != path->mnt)
857 mntput(nd->path.mnt);
859 nd->path.mnt = path->mnt;
860 nd->path.dentry = path->dentry;
863 static int nd_jump_root(struct nameidata *nd)
865 if (nd->flags & LOOKUP_RCU) {
869 nd->inode = d->d_inode;
870 nd->seq = nd->root_seq;
871 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
877 nd->inode = nd->path.dentry->d_inode;
879 nd->flags |= LOOKUP_JUMPED;
884 * Helper to directly jump to a known parsed path from ->get_link,
885 * caller must have taken a reference to path beforehand.
887 void nd_jump_link(struct path *path)
889 struct nameidata *nd = current->nameidata;
893 nd->inode = nd->path.dentry->d_inode;
894 nd->flags |= LOOKUP_JUMPED;
897 static inline void put_link(struct nameidata *nd)
899 struct saved *last = nd->stack + --nd->depth;
900 do_delayed_call(&last->done);
901 if (!(nd->flags & LOOKUP_RCU))
902 path_put(&last->link);
905 int sysctl_protected_symlinks __read_mostly = 0;
906 int sysctl_protected_hardlinks __read_mostly = 0;
909 * may_follow_link - Check symlink following for unsafe situations
910 * @nd: nameidata pathwalk data
912 * In the case of the sysctl_protected_symlinks sysctl being enabled,
913 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
914 * in a sticky world-writable directory. This is to protect privileged
915 * processes from failing races against path names that may change out
916 * from under them by way of other users creating malicious symlinks.
917 * It will permit symlinks to be followed only when outside a sticky
918 * world-writable directory, or when the uid of the symlink and follower
919 * match, or when the directory owner matches the symlink's owner.
921 * Returns 0 if following the symlink is allowed, -ve on error.
923 static inline int may_follow_link(struct nameidata *nd)
925 const struct inode *inode;
926 const struct inode *parent;
929 if (!sysctl_protected_symlinks)
932 /* Allowed if owner and follower match. */
933 inode = nd->link_inode;
934 if (uid_eq(current_cred()->fsuid, inode->i_uid))
937 /* Allowed if parent directory not sticky and world-writable. */
939 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
942 /* Allowed if parent directory and link owner match. */
943 puid = parent->i_uid;
944 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
947 if (nd->flags & LOOKUP_RCU)
950 audit_log_link_denied("follow_link", &nd->stack[0].link);
955 * safe_hardlink_source - Check for safe hardlink conditions
956 * @inode: the source inode to hardlink from
958 * Return false if at least one of the following conditions:
959 * - inode is not a regular file
961 * - inode is setgid and group-exec
962 * - access failure for read and write
964 * Otherwise returns true.
966 static bool safe_hardlink_source(struct inode *inode)
968 umode_t mode = inode->i_mode;
970 /* Special files should not get pinned to the filesystem. */
974 /* Setuid files should not get pinned to the filesystem. */
978 /* Executable setgid files should not get pinned to the filesystem. */
979 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
982 /* Hardlinking to unreadable or unwritable sources is dangerous. */
983 if (inode_permission(inode, MAY_READ | MAY_WRITE))
990 * may_linkat - Check permissions for creating a hardlink
991 * @link: the source to hardlink from
993 * Block hardlink when all of:
994 * - sysctl_protected_hardlinks enabled
995 * - fsuid does not match inode
996 * - hardlink source is unsafe (see safe_hardlink_source() above)
997 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
999 * Returns 0 if successful, -ve on error.
1001 static int may_linkat(struct path *link)
1003 struct inode *inode;
1005 if (!sysctl_protected_hardlinks)
1008 inode = link->dentry->d_inode;
1010 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1011 * otherwise, it must be a safe source.
1013 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1016 audit_log_link_denied("linkat", link);
1020 static __always_inline
1021 const char *get_link(struct nameidata *nd)
1023 struct saved *last = nd->stack + nd->depth - 1;
1024 struct dentry *dentry = last->link.dentry;
1025 struct inode *inode = nd->link_inode;
1029 if (!(nd->flags & LOOKUP_RCU)) {
1030 touch_atime(&last->link);
1032 } else if (atime_needs_update_rcu(&last->link, inode)) {
1033 if (unlikely(unlazy_walk(nd)))
1034 return ERR_PTR(-ECHILD);
1035 touch_atime(&last->link);
1038 error = security_inode_follow_link(dentry, inode,
1039 nd->flags & LOOKUP_RCU);
1040 if (unlikely(error))
1041 return ERR_PTR(error);
1043 nd->last_type = LAST_BIND;
1044 res = inode->i_link;
1046 const char * (*get)(struct dentry *, struct inode *,
1047 struct delayed_call *);
1048 get = inode->i_op->get_link;
1049 if (nd->flags & LOOKUP_RCU) {
1050 res = get(NULL, inode, &last->done);
1051 if (res == ERR_PTR(-ECHILD)) {
1052 if (unlikely(unlazy_walk(nd)))
1053 return ERR_PTR(-ECHILD);
1054 res = get(dentry, inode, &last->done);
1057 res = get(dentry, inode, &last->done);
1059 if (IS_ERR_OR_NULL(res))
1065 if (unlikely(nd_jump_root(nd)))
1066 return ERR_PTR(-ECHILD);
1067 while (unlikely(*++res == '/'))
1076 * follow_up - Find the mountpoint of path's vfsmount
1078 * Given a path, find the mountpoint of its source file system.
1079 * Replace @path with the path of the mountpoint in the parent mount.
1082 * Return 1 if we went up a level and 0 if we were already at the
1085 int follow_up(struct path *path)
1087 struct mount *mnt = real_mount(path->mnt);
1088 struct mount *parent;
1089 struct dentry *mountpoint;
1091 read_seqlock_excl(&mount_lock);
1092 parent = mnt->mnt_parent;
1093 if (parent == mnt) {
1094 read_sequnlock_excl(&mount_lock);
1097 mntget(&parent->mnt);
1098 mountpoint = dget(mnt->mnt_mountpoint);
1099 read_sequnlock_excl(&mount_lock);
1101 path->dentry = mountpoint;
1103 path->mnt = &parent->mnt;
1106 EXPORT_SYMBOL(follow_up);
1109 * Perform an automount
1110 * - return -EISDIR to tell follow_managed() to stop and return the path we
1113 static int follow_automount(struct path *path, struct nameidata *nd,
1116 struct vfsmount *mnt;
1119 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1122 /* We don't want to mount if someone's just doing a stat -
1123 * unless they're stat'ing a directory and appended a '/' to
1126 * We do, however, want to mount if someone wants to open or
1127 * create a file of any type under the mountpoint, wants to
1128 * traverse through the mountpoint or wants to open the
1129 * mounted directory. Also, autofs may mark negative dentries
1130 * as being automount points. These will need the attentions
1131 * of the daemon to instantiate them before they can be used.
1133 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1134 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1135 path->dentry->d_inode)
1138 nd->total_link_count++;
1139 if (nd->total_link_count >= 40)
1142 mnt = path->dentry->d_op->d_automount(path);
1145 * The filesystem is allowed to return -EISDIR here to indicate
1146 * it doesn't want to automount. For instance, autofs would do
1147 * this so that its userspace daemon can mount on this dentry.
1149 * However, we can only permit this if it's a terminal point in
1150 * the path being looked up; if it wasn't then the remainder of
1151 * the path is inaccessible and we should say so.
1153 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1155 return PTR_ERR(mnt);
1158 if (!mnt) /* mount collision */
1161 if (!*need_mntput) {
1162 /* lock_mount() may release path->mnt on error */
1164 *need_mntput = true;
1166 err = finish_automount(mnt, path);
1170 /* Someone else made a mount here whilst we were busy */
1175 path->dentry = dget(mnt->mnt_root);
1184 * Handle a dentry that is managed in some way.
1185 * - Flagged for transit management (autofs)
1186 * - Flagged as mountpoint
1187 * - Flagged as automount point
1189 * This may only be called in refwalk mode.
1191 * Serialization is taken care of in namespace.c
1193 static int follow_managed(struct path *path, struct nameidata *nd)
1195 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1197 bool need_mntput = false;
1200 /* Given that we're not holding a lock here, we retain the value in a
1201 * local variable for each dentry as we look at it so that we don't see
1202 * the components of that value change under us */
1203 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1204 managed &= DCACHE_MANAGED_DENTRY,
1205 unlikely(managed != 0)) {
1206 /* Allow the filesystem to manage the transit without i_mutex
1208 if (managed & DCACHE_MANAGE_TRANSIT) {
1209 BUG_ON(!path->dentry->d_op);
1210 BUG_ON(!path->dentry->d_op->d_manage);
1211 ret = path->dentry->d_op->d_manage(path, false);
1216 /* Transit to a mounted filesystem. */
1217 if (managed & DCACHE_MOUNTED) {
1218 struct vfsmount *mounted = lookup_mnt(path);
1223 path->mnt = mounted;
1224 path->dentry = dget(mounted->mnt_root);
1229 /* Something is mounted on this dentry in another
1230 * namespace and/or whatever was mounted there in this
1231 * namespace got unmounted before lookup_mnt() could
1235 /* Handle an automount point */
1236 if (managed & DCACHE_NEED_AUTOMOUNT) {
1237 ret = follow_automount(path, nd, &need_mntput);
1243 /* We didn't change the current path point */
1247 if (need_mntput && path->mnt == mnt)
1249 if (ret == -EISDIR || !ret)
1252 nd->flags |= LOOKUP_JUMPED;
1253 if (unlikely(ret < 0))
1254 path_put_conditional(path, nd);
1258 int follow_down_one(struct path *path)
1260 struct vfsmount *mounted;
1262 mounted = lookup_mnt(path);
1266 path->mnt = mounted;
1267 path->dentry = dget(mounted->mnt_root);
1272 EXPORT_SYMBOL(follow_down_one);
1274 static inline int managed_dentry_rcu(const struct path *path)
1276 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1277 path->dentry->d_op->d_manage(path, true) : 0;
1281 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1282 * we meet a managed dentry that would need blocking.
1284 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1285 struct inode **inode, unsigned *seqp)
1288 struct mount *mounted;
1290 * Don't forget we might have a non-mountpoint managed dentry
1291 * that wants to block transit.
1293 switch (managed_dentry_rcu(path)) {
1303 if (!d_mountpoint(path->dentry))
1304 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1306 mounted = __lookup_mnt(path->mnt, path->dentry);
1309 path->mnt = &mounted->mnt;
1310 path->dentry = mounted->mnt.mnt_root;
1311 nd->flags |= LOOKUP_JUMPED;
1312 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1314 * Update the inode too. We don't need to re-check the
1315 * dentry sequence number here after this d_inode read,
1316 * because a mount-point is always pinned.
1318 *inode = path->dentry->d_inode;
1320 return !read_seqretry(&mount_lock, nd->m_seq) &&
1321 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1324 static int follow_dotdot_rcu(struct nameidata *nd)
1326 struct inode *inode = nd->inode;
1329 if (path_equal(&nd->path, &nd->root))
1331 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1332 struct dentry *old = nd->path.dentry;
1333 struct dentry *parent = old->d_parent;
1336 inode = parent->d_inode;
1337 seq = read_seqcount_begin(&parent->d_seq);
1338 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1340 nd->path.dentry = parent;
1342 if (unlikely(!path_connected(&nd->path)))
1346 struct mount *mnt = real_mount(nd->path.mnt);
1347 struct mount *mparent = mnt->mnt_parent;
1348 struct dentry *mountpoint = mnt->mnt_mountpoint;
1349 struct inode *inode2 = mountpoint->d_inode;
1350 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1351 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1353 if (&mparent->mnt == nd->path.mnt)
1355 /* we know that mountpoint was pinned */
1356 nd->path.dentry = mountpoint;
1357 nd->path.mnt = &mparent->mnt;
1362 while (unlikely(d_mountpoint(nd->path.dentry))) {
1363 struct mount *mounted;
1364 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1365 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1369 nd->path.mnt = &mounted->mnt;
1370 nd->path.dentry = mounted->mnt.mnt_root;
1371 inode = nd->path.dentry->d_inode;
1372 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1379 * Follow down to the covering mount currently visible to userspace. At each
1380 * point, the filesystem owning that dentry may be queried as to whether the
1381 * caller is permitted to proceed or not.
1383 int follow_down(struct path *path)
1388 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1389 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1390 /* Allow the filesystem to manage the transit without i_mutex
1393 * We indicate to the filesystem if someone is trying to mount
1394 * something here. This gives autofs the chance to deny anyone
1395 * other than its daemon the right to mount on its
1398 * The filesystem may sleep at this point.
1400 if (managed & DCACHE_MANAGE_TRANSIT) {
1401 BUG_ON(!path->dentry->d_op);
1402 BUG_ON(!path->dentry->d_op->d_manage);
1403 ret = path->dentry->d_op->d_manage(path, false);
1405 return ret == -EISDIR ? 0 : ret;
1408 /* Transit to a mounted filesystem. */
1409 if (managed & DCACHE_MOUNTED) {
1410 struct vfsmount *mounted = lookup_mnt(path);
1415 path->mnt = mounted;
1416 path->dentry = dget(mounted->mnt_root);
1420 /* Don't handle automount points here */
1425 EXPORT_SYMBOL(follow_down);
1428 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1430 static void follow_mount(struct path *path)
1432 while (d_mountpoint(path->dentry)) {
1433 struct vfsmount *mounted = lookup_mnt(path);
1438 path->mnt = mounted;
1439 path->dentry = dget(mounted->mnt_root);
1443 static int path_parent_directory(struct path *path)
1445 struct dentry *old = path->dentry;
1446 /* rare case of legitimate dget_parent()... */
1447 path->dentry = dget_parent(path->dentry);
1449 if (unlikely(!path_connected(path)))
1454 static int follow_dotdot(struct nameidata *nd)
1457 if (nd->path.dentry == nd->root.dentry &&
1458 nd->path.mnt == nd->root.mnt) {
1461 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1462 int ret = path_parent_directory(&nd->path);
1467 if (!follow_up(&nd->path))
1470 follow_mount(&nd->path);
1471 nd->inode = nd->path.dentry->d_inode;
1476 * This looks up the name in dcache and possibly revalidates the found dentry.
1477 * NULL is returned if the dentry does not exist in the cache.
1479 static struct dentry *lookup_dcache(const struct qstr *name,
1483 struct dentry *dentry = d_lookup(dir, name);
1485 int error = d_revalidate(dentry, flags);
1486 if (unlikely(error <= 0)) {
1488 d_invalidate(dentry);
1490 return ERR_PTR(error);
1497 * Call i_op->lookup on the dentry. The dentry must be negative and
1500 * dir->d_inode->i_mutex must be held
1502 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1507 /* Don't create child dentry for a dead directory. */
1508 if (unlikely(IS_DEADDIR(dir))) {
1510 return ERR_PTR(-ENOENT);
1513 old = dir->i_op->lookup(dir, dentry, flags);
1514 if (unlikely(old)) {
1521 static struct dentry *__lookup_hash(const struct qstr *name,
1522 struct dentry *base, unsigned int flags)
1524 struct dentry *dentry = lookup_dcache(name, base, flags);
1529 dentry = d_alloc(base, name);
1530 if (unlikely(!dentry))
1531 return ERR_PTR(-ENOMEM);
1533 return lookup_real(base->d_inode, dentry, flags);
1536 static int lookup_fast(struct nameidata *nd,
1537 struct path *path, struct inode **inode,
1540 struct vfsmount *mnt = nd->path.mnt;
1541 struct dentry *dentry, *parent = nd->path.dentry;
1546 * Rename seqlock is not required here because in the off chance
1547 * of a false negative due to a concurrent rename, the caller is
1548 * going to fall back to non-racy lookup.
1550 if (nd->flags & LOOKUP_RCU) {
1553 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1554 if (unlikely(!dentry)) {
1555 if (unlazy_walk(nd))
1561 * This sequence count validates that the inode matches
1562 * the dentry name information from lookup.
1564 *inode = d_backing_inode(dentry);
1565 negative = d_is_negative(dentry);
1566 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1570 * This sequence count validates that the parent had no
1571 * changes while we did the lookup of the dentry above.
1573 * The memory barrier in read_seqcount_begin of child is
1574 * enough, we can use __read_seqcount_retry here.
1576 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1580 status = d_revalidate(dentry, nd->flags);
1581 if (likely(status > 0)) {
1583 * Note: do negative dentry check after revalidation in
1584 * case that drops it.
1586 if (unlikely(negative))
1589 path->dentry = dentry;
1590 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1593 if (unlazy_child(nd, dentry, seq))
1595 if (unlikely(status == -ECHILD))
1596 /* we'd been told to redo it in non-rcu mode */
1597 status = d_revalidate(dentry, nd->flags);
1599 dentry = __d_lookup(parent, &nd->last);
1600 if (unlikely(!dentry))
1602 status = d_revalidate(dentry, nd->flags);
1604 if (unlikely(status <= 0)) {
1606 d_invalidate(dentry);
1610 if (unlikely(d_is_negative(dentry))) {
1616 path->dentry = dentry;
1617 err = follow_managed(path, nd);
1618 if (likely(err > 0))
1619 *inode = d_backing_inode(path->dentry);
1623 /* Fast lookup failed, do it the slow way */
1624 static struct dentry *lookup_slow(const struct qstr *name,
1628 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1629 struct inode *inode = dir->d_inode;
1630 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1632 inode_lock_shared(inode);
1633 /* Don't go there if it's already dead */
1634 if (unlikely(IS_DEADDIR(inode)))
1637 dentry = d_alloc_parallel(dir, name, &wq);
1640 if (unlikely(!d_in_lookup(dentry))) {
1641 if (!(flags & LOOKUP_NO_REVAL)) {
1642 int error = d_revalidate(dentry, flags);
1643 if (unlikely(error <= 0)) {
1645 d_invalidate(dentry);
1650 dentry = ERR_PTR(error);
1654 old = inode->i_op->lookup(inode, dentry, flags);
1655 d_lookup_done(dentry);
1656 if (unlikely(old)) {
1662 inode_unlock_shared(inode);
1666 static inline int may_lookup(struct nameidata *nd)
1668 if (nd->flags & LOOKUP_RCU) {
1669 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1672 if (unlazy_walk(nd))
1675 return inode_permission(nd->inode, MAY_EXEC);
1678 static inline int handle_dots(struct nameidata *nd, int type)
1680 if (type == LAST_DOTDOT) {
1683 if (nd->flags & LOOKUP_RCU) {
1684 return follow_dotdot_rcu(nd);
1686 return follow_dotdot(nd);
1691 static int pick_link(struct nameidata *nd, struct path *link,
1692 struct inode *inode, unsigned seq)
1696 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1697 path_to_nameidata(link, nd);
1700 if (!(nd->flags & LOOKUP_RCU)) {
1701 if (link->mnt == nd->path.mnt)
1704 error = nd_alloc_stack(nd);
1705 if (unlikely(error)) {
1706 if (error == -ECHILD) {
1707 if (unlikely(!legitimize_path(nd, link, seq))) {
1710 nd->flags &= ~LOOKUP_RCU;
1711 nd->path.mnt = NULL;
1712 nd->path.dentry = NULL;
1713 if (!(nd->flags & LOOKUP_ROOT))
1714 nd->root.mnt = NULL;
1716 } else if (likely(unlazy_walk(nd)) == 0)
1717 error = nd_alloc_stack(nd);
1725 last = nd->stack + nd->depth++;
1727 clear_delayed_call(&last->done);
1728 nd->link_inode = inode;
1733 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1736 * Do we need to follow links? We _really_ want to be able
1737 * to do this check without having to look at inode->i_op,
1738 * so we keep a cache of "no, this doesn't need follow_link"
1739 * for the common case.
1741 static inline int step_into(struct nameidata *nd, struct path *path,
1742 int flags, struct inode *inode, unsigned seq)
1744 if (!(flags & WALK_MORE) && nd->depth)
1746 if (likely(!d_is_symlink(path->dentry)) ||
1747 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1748 /* not a symlink or should not follow */
1749 path_to_nameidata(path, nd);
1754 /* make sure that d_is_symlink above matches inode */
1755 if (nd->flags & LOOKUP_RCU) {
1756 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1759 return pick_link(nd, path, inode, seq);
1762 static int walk_component(struct nameidata *nd, int flags)
1765 struct inode *inode;
1769 * "." and ".." are special - ".." especially so because it has
1770 * to be able to know about the current root directory and
1771 * parent relationships.
1773 if (unlikely(nd->last_type != LAST_NORM)) {
1774 err = handle_dots(nd, nd->last_type);
1775 if (!(flags & WALK_MORE) && nd->depth)
1779 err = lookup_fast(nd, &path, &inode, &seq);
1780 if (unlikely(err <= 0)) {
1783 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1785 if (IS_ERR(path.dentry))
1786 return PTR_ERR(path.dentry);
1788 path.mnt = nd->path.mnt;
1789 err = follow_managed(&path, nd);
1790 if (unlikely(err < 0))
1793 if (unlikely(d_is_negative(path.dentry))) {
1794 path_to_nameidata(&path, nd);
1798 seq = 0; /* we are already out of RCU mode */
1799 inode = d_backing_inode(path.dentry);
1802 return step_into(nd, &path, flags, inode, seq);
1806 * We can do the critical dentry name comparison and hashing
1807 * operations one word at a time, but we are limited to:
1809 * - Architectures with fast unaligned word accesses. We could
1810 * do a "get_unaligned()" if this helps and is sufficiently
1813 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1814 * do not trap on the (extremely unlikely) case of a page
1815 * crossing operation.
1817 * - Furthermore, we need an efficient 64-bit compile for the
1818 * 64-bit case in order to generate the "number of bytes in
1819 * the final mask". Again, that could be replaced with a
1820 * efficient population count instruction or similar.
1822 #ifdef CONFIG_DCACHE_WORD_ACCESS
1824 #include <asm/word-at-a-time.h>
1828 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1830 #elif defined(CONFIG_64BIT)
1832 * Register pressure in the mixing function is an issue, particularly
1833 * on 32-bit x86, but almost any function requires one state value and
1834 * one temporary. Instead, use a function designed for two state values
1835 * and no temporaries.
1837 * This function cannot create a collision in only two iterations, so
1838 * we have two iterations to achieve avalanche. In those two iterations,
1839 * we have six layers of mixing, which is enough to spread one bit's
1840 * influence out to 2^6 = 64 state bits.
1842 * Rotate constants are scored by considering either 64 one-bit input
1843 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1844 * probability of that delta causing a change to each of the 128 output
1845 * bits, using a sample of random initial states.
1847 * The Shannon entropy of the computed probabilities is then summed
1848 * to produce a score. Ideally, any input change has a 50% chance of
1849 * toggling any given output bit.
1851 * Mixing scores (in bits) for (12,45):
1852 * Input delta: 1-bit 2-bit
1853 * 1 round: 713.3 42542.6
1854 * 2 rounds: 2753.7 140389.8
1855 * 3 rounds: 5954.1 233458.2
1856 * 4 rounds: 7862.6 256672.2
1857 * Perfect: 8192 258048
1858 * (64*128) (64*63/2 * 128)
1860 #define HASH_MIX(x, y, a) \
1862 y ^= x, x = rol64(x,12),\
1863 x += y, y = rol64(y,45),\
1867 * Fold two longs into one 32-bit hash value. This must be fast, but
1868 * latency isn't quite as critical, as there is a fair bit of additional
1869 * work done before the hash value is used.
1871 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1873 y ^= x * GOLDEN_RATIO_64;
1874 y *= GOLDEN_RATIO_64;
1878 #else /* 32-bit case */
1881 * Mixing scores (in bits) for (7,20):
1882 * Input delta: 1-bit 2-bit
1883 * 1 round: 330.3 9201.6
1884 * 2 rounds: 1246.4 25475.4
1885 * 3 rounds: 1907.1 31295.1
1886 * 4 rounds: 2042.3 31718.6
1887 * Perfect: 2048 31744
1888 * (32*64) (32*31/2 * 64)
1890 #define HASH_MIX(x, y, a) \
1892 y ^= x, x = rol32(x, 7),\
1893 x += y, y = rol32(y,20),\
1896 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1898 /* Use arch-optimized multiply if one exists */
1899 return __hash_32(y ^ __hash_32(x));
1905 * Return the hash of a string of known length. This is carfully
1906 * designed to match hash_name(), which is the more critical function.
1907 * In particular, we must end by hashing a final word containing 0..7
1908 * payload bytes, to match the way that hash_name() iterates until it
1909 * finds the delimiter after the name.
1911 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1913 unsigned long a, x = 0, y = (unsigned long)salt;
1918 a = load_unaligned_zeropad(name);
1919 if (len < sizeof(unsigned long))
1922 name += sizeof(unsigned long);
1923 len -= sizeof(unsigned long);
1925 x ^= a & bytemask_from_count(len);
1927 return fold_hash(x, y);
1929 EXPORT_SYMBOL(full_name_hash);
1931 /* Return the "hash_len" (hash and length) of a null-terminated string */
1932 u64 hashlen_string(const void *salt, const char *name)
1934 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1935 unsigned long adata, mask, len;
1936 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1943 len += sizeof(unsigned long);
1945 a = load_unaligned_zeropad(name+len);
1946 } while (!has_zero(a, &adata, &constants));
1948 adata = prep_zero_mask(a, adata, &constants);
1949 mask = create_zero_mask(adata);
1950 x ^= a & zero_bytemask(mask);
1952 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1954 EXPORT_SYMBOL(hashlen_string);
1957 * Calculate the length and hash of the path component, and
1958 * return the "hash_len" as the result.
1960 static inline u64 hash_name(const void *salt, const char *name)
1962 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1963 unsigned long adata, bdata, mask, len;
1964 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1971 len += sizeof(unsigned long);
1973 a = load_unaligned_zeropad(name+len);
1974 b = a ^ REPEAT_BYTE('/');
1975 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1977 adata = prep_zero_mask(a, adata, &constants);
1978 bdata = prep_zero_mask(b, bdata, &constants);
1979 mask = create_zero_mask(adata | bdata);
1980 x ^= a & zero_bytemask(mask);
1982 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1985 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1987 /* Return the hash of a string of known length */
1988 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1990 unsigned long hash = init_name_hash(salt);
1992 hash = partial_name_hash((unsigned char)*name++, hash);
1993 return end_name_hash(hash);
1995 EXPORT_SYMBOL(full_name_hash);
1997 /* Return the "hash_len" (hash and length) of a null-terminated string */
1998 u64 hashlen_string(const void *salt, const char *name)
2000 unsigned long hash = init_name_hash(salt);
2001 unsigned long len = 0, c;
2003 c = (unsigned char)*name;
2006 hash = partial_name_hash(c, hash);
2007 c = (unsigned char)name[len];
2009 return hashlen_create(end_name_hash(hash), len);
2011 EXPORT_SYMBOL(hashlen_string);
2014 * We know there's a real path component here of at least
2017 static inline u64 hash_name(const void *salt, const char *name)
2019 unsigned long hash = init_name_hash(salt);
2020 unsigned long len = 0, c;
2022 c = (unsigned char)*name;
2025 hash = partial_name_hash(c, hash);
2026 c = (unsigned char)name[len];
2027 } while (c && c != '/');
2028 return hashlen_create(end_name_hash(hash), len);
2035 * This is the basic name resolution function, turning a pathname into
2036 * the final dentry. We expect 'base' to be positive and a directory.
2038 * Returns 0 and nd will have valid dentry and mnt on success.
2039 * Returns error and drops reference to input namei data on failure.
2041 static int link_path_walk(const char *name, struct nameidata *nd)
2050 /* At this point we know we have a real path component. */
2055 err = may_lookup(nd);
2059 hash_len = hash_name(nd->path.dentry, name);
2062 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2064 if (name[1] == '.') {
2066 nd->flags |= LOOKUP_JUMPED;
2072 if (likely(type == LAST_NORM)) {
2073 struct dentry *parent = nd->path.dentry;
2074 nd->flags &= ~LOOKUP_JUMPED;
2075 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2076 struct qstr this = { { .hash_len = hash_len }, .name = name };
2077 err = parent->d_op->d_hash(parent, &this);
2080 hash_len = this.hash_len;
2085 nd->last.hash_len = hash_len;
2086 nd->last.name = name;
2087 nd->last_type = type;
2089 name += hashlen_len(hash_len);
2093 * If it wasn't NUL, we know it was '/'. Skip that
2094 * slash, and continue until no more slashes.
2098 } while (unlikely(*name == '/'));
2099 if (unlikely(!*name)) {
2101 /* pathname body, done */
2104 name = nd->stack[nd->depth - 1].name;
2105 /* trailing symlink, done */
2108 /* last component of nested symlink */
2109 err = walk_component(nd, WALK_FOLLOW);
2111 /* not the last component */
2112 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2118 const char *s = get_link(nd);
2127 nd->stack[nd->depth - 1].name = name;
2132 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2133 if (nd->flags & LOOKUP_RCU) {
2134 if (unlazy_walk(nd))
2142 static const char *path_init(struct nameidata *nd, unsigned flags)
2144 const char *s = nd->name->name;
2147 flags &= ~LOOKUP_RCU;
2149 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2150 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2152 if (flags & LOOKUP_ROOT) {
2153 struct dentry *root = nd->root.dentry;
2154 struct inode *inode = root->d_inode;
2155 if (*s && unlikely(!d_can_lookup(root)))
2156 return ERR_PTR(-ENOTDIR);
2157 nd->path = nd->root;
2159 if (flags & LOOKUP_RCU) {
2161 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2162 nd->root_seq = nd->seq;
2163 nd->m_seq = read_seqbegin(&mount_lock);
2165 path_get(&nd->path);
2170 nd->root.mnt = NULL;
2171 nd->path.mnt = NULL;
2172 nd->path.dentry = NULL;
2174 nd->m_seq = read_seqbegin(&mount_lock);
2176 if (flags & LOOKUP_RCU)
2179 if (likely(!nd_jump_root(nd)))
2181 nd->root.mnt = NULL;
2183 return ERR_PTR(-ECHILD);
2184 } else if (nd->dfd == AT_FDCWD) {
2185 if (flags & LOOKUP_RCU) {
2186 struct fs_struct *fs = current->fs;
2192 seq = read_seqcount_begin(&fs->seq);
2194 nd->inode = nd->path.dentry->d_inode;
2195 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2196 } while (read_seqcount_retry(&fs->seq, seq));
2198 get_fs_pwd(current->fs, &nd->path);
2199 nd->inode = nd->path.dentry->d_inode;
2203 /* Caller must check execute permissions on the starting path component */
2204 struct fd f = fdget_raw(nd->dfd);
2205 struct dentry *dentry;
2208 return ERR_PTR(-EBADF);
2210 dentry = f.file->f_path.dentry;
2213 if (!d_can_lookup(dentry)) {
2215 return ERR_PTR(-ENOTDIR);
2219 nd->path = f.file->f_path;
2220 if (flags & LOOKUP_RCU) {
2222 nd->inode = nd->path.dentry->d_inode;
2223 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2225 path_get(&nd->path);
2226 nd->inode = nd->path.dentry->d_inode;
2233 static const char *trailing_symlink(struct nameidata *nd)
2236 int error = may_follow_link(nd);
2237 if (unlikely(error))
2238 return ERR_PTR(error);
2239 nd->flags |= LOOKUP_PARENT;
2240 nd->stack[0].name = NULL;
2245 static inline int lookup_last(struct nameidata *nd)
2247 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2248 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2250 nd->flags &= ~LOOKUP_PARENT;
2251 return walk_component(nd, 0);
2254 static int handle_lookup_down(struct nameidata *nd)
2256 struct path path = nd->path;
2257 struct inode *inode = nd->inode;
2258 unsigned seq = nd->seq;
2261 if (nd->flags & LOOKUP_RCU) {
2263 * don't bother with unlazy_walk on failure - we are
2264 * at the very beginning of walk, so we lose nothing
2265 * if we simply redo everything in non-RCU mode
2267 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2271 err = follow_managed(&path, nd);
2272 if (unlikely(err < 0))
2274 inode = d_backing_inode(path.dentry);
2277 path_to_nameidata(&path, nd);
2283 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2284 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2286 const char *s = path_init(nd, flags);
2292 if (unlikely(flags & LOOKUP_DOWN)) {
2293 err = handle_lookup_down(nd);
2294 if (unlikely(err < 0)) {
2300 while (!(err = link_path_walk(s, nd))
2301 && ((err = lookup_last(nd)) > 0)) {
2302 s = trailing_symlink(nd);
2309 err = complete_walk(nd);
2311 if (!err && nd->flags & LOOKUP_DIRECTORY)
2312 if (!d_can_lookup(nd->path.dentry))
2316 nd->path.mnt = NULL;
2317 nd->path.dentry = NULL;
2323 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2324 struct path *path, struct path *root)
2327 struct nameidata nd;
2329 return PTR_ERR(name);
2330 if (unlikely(root)) {
2332 flags |= LOOKUP_ROOT;
2334 set_nameidata(&nd, dfd, name);
2335 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2336 if (unlikely(retval == -ECHILD))
2337 retval = path_lookupat(&nd, flags, path);
2338 if (unlikely(retval == -ESTALE))
2339 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2341 if (likely(!retval))
2342 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2343 restore_nameidata();
2348 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2349 static int path_parentat(struct nameidata *nd, unsigned flags,
2350 struct path *parent)
2352 const char *s = path_init(nd, flags);
2356 err = link_path_walk(s, nd);
2358 err = complete_walk(nd);
2361 nd->path.mnt = NULL;
2362 nd->path.dentry = NULL;
2368 static struct filename *filename_parentat(int dfd, struct filename *name,
2369 unsigned int flags, struct path *parent,
2370 struct qstr *last, int *type)
2373 struct nameidata nd;
2377 set_nameidata(&nd, dfd, name);
2378 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2379 if (unlikely(retval == -ECHILD))
2380 retval = path_parentat(&nd, flags, parent);
2381 if (unlikely(retval == -ESTALE))
2382 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2383 if (likely(!retval)) {
2385 *type = nd.last_type;
2386 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2389 name = ERR_PTR(retval);
2391 restore_nameidata();
2395 /* does lookup, returns the object with parent locked */
2396 struct dentry *kern_path_locked(const char *name, struct path *path)
2398 struct filename *filename;
2403 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2405 if (IS_ERR(filename))
2406 return ERR_CAST(filename);
2407 if (unlikely(type != LAST_NORM)) {
2410 return ERR_PTR(-EINVAL);
2412 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2413 d = __lookup_hash(&last, path->dentry, 0);
2415 inode_unlock(path->dentry->d_inode);
2422 int kern_path(const char *name, unsigned int flags, struct path *path)
2424 return filename_lookup(AT_FDCWD, getname_kernel(name),
2427 EXPORT_SYMBOL(kern_path);
2430 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2431 * @dentry: pointer to dentry of the base directory
2432 * @mnt: pointer to vfs mount of the base directory
2433 * @name: pointer to file name
2434 * @flags: lookup flags
2435 * @path: pointer to struct path to fill
2437 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2438 const char *name, unsigned int flags,
2441 struct path root = {.mnt = mnt, .dentry = dentry};
2442 /* the first argument of filename_lookup() is ignored with root */
2443 return filename_lookup(AT_FDCWD, getname_kernel(name),
2444 flags , path, &root);
2446 EXPORT_SYMBOL(vfs_path_lookup);
2449 * lookup_one_len - filesystem helper to lookup single pathname component
2450 * @name: pathname component to lookup
2451 * @base: base directory to lookup from
2452 * @len: maximum length @len should be interpreted to
2454 * Note that this routine is purely a helper for filesystem usage and should
2455 * not be called by generic code.
2457 * The caller must hold base->i_mutex.
2459 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2465 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2469 this.hash = full_name_hash(base, name, len);
2471 return ERR_PTR(-EACCES);
2473 if (unlikely(name[0] == '.')) {
2474 if (len < 2 || (len == 2 && name[1] == '.'))
2475 return ERR_PTR(-EACCES);
2479 c = *(const unsigned char *)name++;
2480 if (c == '/' || c == '\0')
2481 return ERR_PTR(-EACCES);
2484 * See if the low-level filesystem might want
2485 * to use its own hash..
2487 if (base->d_flags & DCACHE_OP_HASH) {
2488 int err = base->d_op->d_hash(base, &this);
2490 return ERR_PTR(err);
2493 err = inode_permission(base->d_inode, MAY_EXEC);
2495 return ERR_PTR(err);
2497 return __lookup_hash(&this, base, 0);
2499 EXPORT_SYMBOL(lookup_one_len);
2502 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2503 * @name: pathname component to lookup
2504 * @base: base directory to lookup from
2505 * @len: maximum length @len should be interpreted to
2507 * Note that this routine is purely a helper for filesystem usage and should
2508 * not be called by generic code.
2510 * Unlike lookup_one_len, it should be called without the parent
2511 * i_mutex held, and will take the i_mutex itself if necessary.
2513 struct dentry *lookup_one_len_unlocked(const char *name,
2514 struct dentry *base, int len)
2523 this.hash = full_name_hash(base, name, len);
2525 return ERR_PTR(-EACCES);
2527 if (unlikely(name[0] == '.')) {
2528 if (len < 2 || (len == 2 && name[1] == '.'))
2529 return ERR_PTR(-EACCES);
2533 c = *(const unsigned char *)name++;
2534 if (c == '/' || c == '\0')
2535 return ERR_PTR(-EACCES);
2538 * See if the low-level filesystem might want
2539 * to use its own hash..
2541 if (base->d_flags & DCACHE_OP_HASH) {
2542 int err = base->d_op->d_hash(base, &this);
2544 return ERR_PTR(err);
2547 err = inode_permission(base->d_inode, MAY_EXEC);
2549 return ERR_PTR(err);
2551 ret = lookup_dcache(&this, base, 0);
2553 ret = lookup_slow(&this, base, 0);
2556 EXPORT_SYMBOL(lookup_one_len_unlocked);
2558 #ifdef CONFIG_UNIX98_PTYS
2559 int path_pts(struct path *path)
2561 /* Find something mounted on "pts" in the same directory as
2564 struct dentry *child, *parent;
2568 ret = path_parent_directory(path);
2572 parent = path->dentry;
2575 child = d_hash_and_lookup(parent, &this);
2579 path->dentry = child;
2586 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2587 struct path *path, int *empty)
2589 return filename_lookup(dfd, getname_flags(name, flags, empty),
2592 EXPORT_SYMBOL(user_path_at_empty);
2595 * mountpoint_last - look up last component for umount
2596 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2598 * This is a special lookup_last function just for umount. In this case, we
2599 * need to resolve the path without doing any revalidation.
2601 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2602 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2603 * in almost all cases, this lookup will be served out of the dcache. The only
2604 * cases where it won't are if nd->last refers to a symlink or the path is
2605 * bogus and it doesn't exist.
2608 * -error: if there was an error during lookup. This includes -ENOENT if the
2609 * lookup found a negative dentry.
2611 * 0: if we successfully resolved nd->last and found it to not to be a
2612 * symlink that needs to be followed.
2614 * 1: if we successfully resolved nd->last and found it to be a symlink
2615 * that needs to be followed.
2618 mountpoint_last(struct nameidata *nd)
2621 struct dentry *dir = nd->path.dentry;
2624 /* If we're in rcuwalk, drop out of it to handle last component */
2625 if (nd->flags & LOOKUP_RCU) {
2626 if (unlazy_walk(nd))
2630 nd->flags &= ~LOOKUP_PARENT;
2632 if (unlikely(nd->last_type != LAST_NORM)) {
2633 error = handle_dots(nd, nd->last_type);
2636 path.dentry = dget(nd->path.dentry);
2638 path.dentry = d_lookup(dir, &nd->last);
2641 * No cached dentry. Mounted dentries are pinned in the
2642 * cache, so that means that this dentry is probably
2643 * a symlink or the path doesn't actually point
2644 * to a mounted dentry.
2646 path.dentry = lookup_slow(&nd->last, dir,
2647 nd->flags | LOOKUP_NO_REVAL);
2648 if (IS_ERR(path.dentry))
2649 return PTR_ERR(path.dentry);
2652 if (d_is_negative(path.dentry)) {
2656 path.mnt = nd->path.mnt;
2657 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2661 * path_mountpoint - look up a path to be umounted
2662 * @nd: lookup context
2663 * @flags: lookup flags
2664 * @path: pointer to container for result
2666 * Look up the given name, but don't attempt to revalidate the last component.
2667 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2670 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2672 const char *s = path_init(nd, flags);
2676 while (!(err = link_path_walk(s, nd)) &&
2677 (err = mountpoint_last(nd)) > 0) {
2678 s = trailing_symlink(nd);
2686 nd->path.mnt = NULL;
2687 nd->path.dentry = NULL;
2695 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2698 struct nameidata nd;
2701 return PTR_ERR(name);
2702 set_nameidata(&nd, dfd, name);
2703 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2704 if (unlikely(error == -ECHILD))
2705 error = path_mountpoint(&nd, flags, path);
2706 if (unlikely(error == -ESTALE))
2707 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2709 audit_inode(name, path->dentry, 0);
2710 restore_nameidata();
2716 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2717 * @dfd: directory file descriptor
2718 * @name: pathname from userland
2719 * @flags: lookup flags
2720 * @path: pointer to container to hold result
2722 * A umount is a special case for path walking. We're not actually interested
2723 * in the inode in this situation, and ESTALE errors can be a problem. We
2724 * simply want track down the dentry and vfsmount attached at the mountpoint
2725 * and avoid revalidating the last component.
2727 * Returns 0 and populates "path" on success.
2730 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2733 return filename_mountpoint(dfd, getname(name), path, flags);
2737 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2740 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2742 EXPORT_SYMBOL(kern_path_mountpoint);
2744 int __check_sticky(struct inode *dir, struct inode *inode)
2746 kuid_t fsuid = current_fsuid();
2748 if (uid_eq(inode->i_uid, fsuid))
2750 if (uid_eq(dir->i_uid, fsuid))
2752 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2754 EXPORT_SYMBOL(__check_sticky);
2757 * Check whether we can remove a link victim from directory dir, check
2758 * whether the type of victim is right.
2759 * 1. We can't do it if dir is read-only (done in permission())
2760 * 2. We should have write and exec permissions on dir
2761 * 3. We can't remove anything from append-only dir
2762 * 4. We can't do anything with immutable dir (done in permission())
2763 * 5. If the sticky bit on dir is set we should either
2764 * a. be owner of dir, or
2765 * b. be owner of victim, or
2766 * c. have CAP_FOWNER capability
2767 * 6. If the victim is append-only or immutable we can't do antyhing with
2768 * links pointing to it.
2769 * 7. If the victim has an unknown uid or gid we can't change the inode.
2770 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2771 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2772 * 10. We can't remove a root or mountpoint.
2773 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2774 * nfs_async_unlink().
2776 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2778 struct inode *inode = d_backing_inode(victim);
2781 if (d_is_negative(victim))
2785 BUG_ON(victim->d_parent->d_inode != dir);
2786 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2788 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2794 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2795 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2798 if (!d_is_dir(victim))
2800 if (IS_ROOT(victim))
2802 } else if (d_is_dir(victim))
2804 if (IS_DEADDIR(dir))
2806 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2811 /* Check whether we can create an object with dentry child in directory
2813 * 1. We can't do it if child already exists (open has special treatment for
2814 * this case, but since we are inlined it's OK)
2815 * 2. We can't do it if dir is read-only (done in permission())
2816 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2817 * 4. We should have write and exec permissions on dir
2818 * 5. We can't do it if dir is immutable (done in permission())
2820 static inline int may_create(struct inode *dir, struct dentry *child)
2822 struct user_namespace *s_user_ns;
2823 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2826 if (IS_DEADDIR(dir))
2828 s_user_ns = dir->i_sb->s_user_ns;
2829 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2830 !kgid_has_mapping(s_user_ns, current_fsgid()))
2832 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2836 * p1 and p2 should be directories on the same fs.
2838 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2843 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2847 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2849 p = d_ancestor(p2, p1);
2851 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2852 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2856 p = d_ancestor(p1, p2);
2858 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2859 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2863 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2864 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2867 EXPORT_SYMBOL(lock_rename);
2869 void unlock_rename(struct dentry *p1, struct dentry *p2)
2871 inode_unlock(p1->d_inode);
2873 inode_unlock(p2->d_inode);
2874 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2877 EXPORT_SYMBOL(unlock_rename);
2879 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2882 int error = may_create(dir, dentry);
2886 if (!dir->i_op->create)
2887 return -EACCES; /* shouldn't it be ENOSYS? */
2890 error = security_inode_create(dir, dentry, mode);
2893 error = dir->i_op->create(dir, dentry, mode, want_excl);
2895 fsnotify_create(dir, dentry);
2898 EXPORT_SYMBOL(vfs_create);
2900 bool may_open_dev(const struct path *path)
2902 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2903 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2906 static int may_open(const struct path *path, int acc_mode, int flag)
2908 struct dentry *dentry = path->dentry;
2909 struct inode *inode = dentry->d_inode;
2915 switch (inode->i_mode & S_IFMT) {
2919 if (acc_mode & MAY_WRITE)
2924 if (!may_open_dev(path))
2933 error = inode_permission(inode, MAY_OPEN | acc_mode);
2938 * An append-only file must be opened in append mode for writing.
2940 if (IS_APPEND(inode)) {
2941 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2947 /* O_NOATIME can only be set by the owner or superuser */
2948 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2954 static int handle_truncate(struct file *filp)
2956 const struct path *path = &filp->f_path;
2957 struct inode *inode = path->dentry->d_inode;
2958 int error = get_write_access(inode);
2962 * Refuse to truncate files with mandatory locks held on them.
2964 error = locks_verify_locked(filp);
2966 error = security_path_truncate(path);
2968 error = do_truncate(path->dentry, 0,
2969 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2972 put_write_access(inode);
2976 static inline int open_to_namei_flags(int flag)
2978 if ((flag & O_ACCMODE) == 3)
2983 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2985 struct user_namespace *s_user_ns;
2986 int error = security_path_mknod(dir, dentry, mode, 0);
2990 s_user_ns = dir->dentry->d_sb->s_user_ns;
2991 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2992 !kgid_has_mapping(s_user_ns, current_fsgid()))
2995 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2999 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3003 * Attempt to atomically look up, create and open a file from a negative
3006 * Returns 0 if successful. The file will have been created and attached to
3007 * @file by the filesystem calling finish_open().
3009 * Returns 1 if the file was looked up only or didn't need creating. The
3010 * caller will need to perform the open themselves. @path will have been
3011 * updated to point to the new dentry. This may be negative.
3013 * Returns an error code otherwise.
3015 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3016 struct path *path, struct file *file,
3017 const struct open_flags *op,
3018 int open_flag, umode_t mode,
3021 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3022 struct inode *dir = nd->path.dentry->d_inode;
3025 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3026 open_flag &= ~O_TRUNC;
3028 if (nd->flags & LOOKUP_DIRECTORY)
3029 open_flag |= O_DIRECTORY;
3031 file->f_path.dentry = DENTRY_NOT_SET;
3032 file->f_path.mnt = nd->path.mnt;
3033 error = dir->i_op->atomic_open(dir, dentry, file,
3034 open_to_namei_flags(open_flag),
3036 d_lookup_done(dentry);
3039 * We didn't have the inode before the open, so check open
3042 int acc_mode = op->acc_mode;
3043 if (*opened & FILE_CREATED) {
3044 WARN_ON(!(open_flag & O_CREAT));
3045 fsnotify_create(dir, dentry);
3048 error = may_open(&file->f_path, acc_mode, open_flag);
3049 if (WARN_ON(error > 0))
3051 } else if (error > 0) {
3052 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3055 if (file->f_path.dentry) {
3057 dentry = file->f_path.dentry;
3059 if (*opened & FILE_CREATED)
3060 fsnotify_create(dir, dentry);
3061 if (unlikely(d_is_negative(dentry))) {
3064 path->dentry = dentry;
3065 path->mnt = nd->path.mnt;
3075 * Look up and maybe create and open the last component.
3077 * Must be called with i_mutex held on parent.
3079 * Returns 0 if the file was successfully atomically created (if necessary) and
3080 * opened. In this case the file will be returned attached to @file.
3082 * Returns 1 if the file was not completely opened at this time, though lookups
3083 * and creations will have been performed and the dentry returned in @path will
3084 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3085 * specified then a negative dentry may be returned.
3087 * An error code is returned otherwise.
3089 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3090 * cleared otherwise prior to returning.
3092 static int lookup_open(struct nameidata *nd, struct path *path,
3094 const struct open_flags *op,
3095 bool got_write, int *opened)
3097 struct dentry *dir = nd->path.dentry;
3098 struct inode *dir_inode = dir->d_inode;
3099 int open_flag = op->open_flag;
3100 struct dentry *dentry;
3101 int error, create_error = 0;
3102 umode_t mode = op->mode;
3103 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3105 if (unlikely(IS_DEADDIR(dir_inode)))
3108 *opened &= ~FILE_CREATED;
3109 dentry = d_lookup(dir, &nd->last);
3112 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3114 return PTR_ERR(dentry);
3116 if (d_in_lookup(dentry))
3119 error = d_revalidate(dentry, nd->flags);
3120 if (likely(error > 0))
3124 d_invalidate(dentry);
3128 if (dentry->d_inode) {
3129 /* Cached positive dentry: will open in f_op->open */
3134 * Checking write permission is tricky, bacuse we don't know if we are
3135 * going to actually need it: O_CREAT opens should work as long as the
3136 * file exists. But checking existence breaks atomicity. The trick is
3137 * to check access and if not granted clear O_CREAT from the flags.
3139 * Another problem is returing the "right" error value (e.g. for an
3140 * O_EXCL open we want to return EEXIST not EROFS).
3142 if (open_flag & O_CREAT) {
3143 if (!IS_POSIXACL(dir->d_inode))
3144 mode &= ~current_umask();
3145 if (unlikely(!got_write)) {
3146 create_error = -EROFS;
3147 open_flag &= ~O_CREAT;
3148 if (open_flag & (O_EXCL | O_TRUNC))
3150 /* No side effects, safe to clear O_CREAT */
3152 create_error = may_o_create(&nd->path, dentry, mode);
3154 open_flag &= ~O_CREAT;
3155 if (open_flag & O_EXCL)
3159 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3160 unlikely(!got_write)) {
3162 * No O_CREATE -> atomicity not a requirement -> fall
3163 * back to lookup + open
3168 if (dir_inode->i_op->atomic_open) {
3169 error = atomic_open(nd, dentry, path, file, op, open_flag,
3171 if (unlikely(error == -ENOENT) && create_error)
3172 error = create_error;
3177 if (d_in_lookup(dentry)) {
3178 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3180 d_lookup_done(dentry);
3181 if (unlikely(res)) {
3183 error = PTR_ERR(res);
3191 /* Negative dentry, just create the file */
3192 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3193 *opened |= FILE_CREATED;
3194 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3195 if (!dir_inode->i_op->create) {
3199 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3200 open_flag & O_EXCL);
3203 fsnotify_create(dir_inode, dentry);
3205 if (unlikely(create_error) && !dentry->d_inode) {
3206 error = create_error;
3210 path->dentry = dentry;
3211 path->mnt = nd->path.mnt;
3220 * Handle the last step of open()
3222 static int do_last(struct nameidata *nd,
3223 struct file *file, const struct open_flags *op,
3226 struct dentry *dir = nd->path.dentry;
3227 int open_flag = op->open_flag;
3228 bool will_truncate = (open_flag & O_TRUNC) != 0;
3229 bool got_write = false;
3230 int acc_mode = op->acc_mode;
3232 struct inode *inode;
3236 nd->flags &= ~LOOKUP_PARENT;
3237 nd->flags |= op->intent;
3239 if (nd->last_type != LAST_NORM) {
3240 error = handle_dots(nd, nd->last_type);
3241 if (unlikely(error))
3246 if (!(open_flag & O_CREAT)) {
3247 if (nd->last.name[nd->last.len])
3248 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3249 /* we _can_ be in RCU mode here */
3250 error = lookup_fast(nd, &path, &inode, &seq);
3251 if (likely(error > 0))
3257 BUG_ON(nd->inode != dir->d_inode);
3258 BUG_ON(nd->flags & LOOKUP_RCU);
3260 /* create side of things */
3262 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3263 * has been cleared when we got to the last component we are
3266 error = complete_walk(nd);
3270 audit_inode(nd->name, dir, LOOKUP_PARENT);
3271 /* trailing slashes? */
3272 if (unlikely(nd->last.name[nd->last.len]))
3276 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3277 error = mnt_want_write(nd->path.mnt);
3281 * do _not_ fail yet - we might not need that or fail with
3282 * a different error; let lookup_open() decide; we'll be
3283 * dropping this one anyway.
3286 if (open_flag & O_CREAT)
3287 inode_lock(dir->d_inode);
3289 inode_lock_shared(dir->d_inode);
3290 error = lookup_open(nd, &path, file, op, got_write, opened);
3291 if (open_flag & O_CREAT)
3292 inode_unlock(dir->d_inode);
3294 inode_unlock_shared(dir->d_inode);
3300 if ((*opened & FILE_CREATED) ||
3301 !S_ISREG(file_inode(file)->i_mode))
3302 will_truncate = false;
3304 audit_inode(nd->name, file->f_path.dentry, 0);
3308 if (*opened & FILE_CREATED) {
3309 /* Don't check for write permission, don't truncate */
3310 open_flag &= ~O_TRUNC;
3311 will_truncate = false;
3313 path_to_nameidata(&path, nd);
3314 goto finish_open_created;
3318 * If atomic_open() acquired write access it is dropped now due to
3319 * possible mount and symlink following (this might be optimized away if
3323 mnt_drop_write(nd->path.mnt);
3327 error = follow_managed(&path, nd);
3328 if (unlikely(error < 0))
3331 if (unlikely(d_is_negative(path.dentry))) {
3332 path_to_nameidata(&path, nd);
3337 * create/update audit record if it already exists.
3339 audit_inode(nd->name, path.dentry, 0);
3341 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3342 path_to_nameidata(&path, nd);
3346 seq = 0; /* out of RCU mode, so the value doesn't matter */
3347 inode = d_backing_inode(path.dentry);
3349 error = step_into(nd, &path, 0, inode, seq);
3350 if (unlikely(error))
3353 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3354 error = complete_walk(nd);
3357 audit_inode(nd->name, nd->path.dentry, 0);
3359 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3362 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3364 if (!d_is_reg(nd->path.dentry))
3365 will_truncate = false;
3367 if (will_truncate) {
3368 error = mnt_want_write(nd->path.mnt);
3373 finish_open_created:
3374 error = may_open(&nd->path, acc_mode, open_flag);
3377 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3378 error = vfs_open(&nd->path, file, current_cred());
3381 *opened |= FILE_OPENED;
3383 error = open_check_o_direct(file);
3385 error = ima_file_check(file, op->acc_mode, *opened);
3386 if (!error && will_truncate)
3387 error = handle_truncate(file);
3389 if (unlikely(error) && (*opened & FILE_OPENED))
3391 if (unlikely(error > 0)) {
3396 mnt_drop_write(nd->path.mnt);
3400 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3402 struct dentry *child = NULL;
3403 struct inode *dir = dentry->d_inode;
3404 struct inode *inode;
3407 /* we want directory to be writable */
3408 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3411 error = -EOPNOTSUPP;
3412 if (!dir->i_op->tmpfile)
3415 child = d_alloc(dentry, &slash_name);
3416 if (unlikely(!child))
3418 error = dir->i_op->tmpfile(dir, child, mode);
3422 inode = child->d_inode;
3423 if (unlikely(!inode))
3425 if (!(open_flag & O_EXCL)) {
3426 spin_lock(&inode->i_lock);
3427 inode->i_state |= I_LINKABLE;
3428 spin_unlock(&inode->i_lock);
3434 return ERR_PTR(error);
3436 EXPORT_SYMBOL(vfs_tmpfile);
3438 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3439 const struct open_flags *op,
3440 struct file *file, int *opened)
3442 struct dentry *child;
3444 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3445 if (unlikely(error))
3447 error = mnt_want_write(path.mnt);
3448 if (unlikely(error))
3450 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3451 error = PTR_ERR(child);
3452 if (unlikely(IS_ERR(child)))
3455 path.dentry = child;
3456 audit_inode(nd->name, child, 0);
3457 /* Don't check for other permissions, the inode was just created */
3458 error = may_open(&path, 0, op->open_flag);
3461 file->f_path.mnt = path.mnt;
3462 error = finish_open(file, child, NULL, opened);
3465 error = open_check_o_direct(file);
3469 mnt_drop_write(path.mnt);
3475 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3478 int error = path_lookupat(nd, flags, &path);
3480 audit_inode(nd->name, path.dentry, 0);
3481 error = vfs_open(&path, file, current_cred());
3487 static struct file *path_openat(struct nameidata *nd,
3488 const struct open_flags *op, unsigned flags)
3495 file = get_empty_filp();
3499 file->f_flags = op->open_flag;
3501 if (unlikely(file->f_flags & __O_TMPFILE)) {
3502 error = do_tmpfile(nd, flags, op, file, &opened);
3506 if (unlikely(file->f_flags & O_PATH)) {
3507 error = do_o_path(nd, flags, file);
3509 opened |= FILE_OPENED;
3513 s = path_init(nd, flags);
3518 while (!(error = link_path_walk(s, nd)) &&
3519 (error = do_last(nd, file, op, &opened)) > 0) {
3520 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3521 s = trailing_symlink(nd);
3529 if (!(opened & FILE_OPENED)) {
3533 if (unlikely(error)) {
3534 if (error == -EOPENSTALE) {
3535 if (flags & LOOKUP_RCU)
3540 file = ERR_PTR(error);
3545 struct file *do_filp_open(int dfd, struct filename *pathname,
3546 const struct open_flags *op)
3548 struct nameidata nd;
3549 int flags = op->lookup_flags;
3552 set_nameidata(&nd, dfd, pathname);
3553 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3554 if (unlikely(filp == ERR_PTR(-ECHILD)))
3555 filp = path_openat(&nd, op, flags);
3556 if (unlikely(filp == ERR_PTR(-ESTALE)))
3557 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3558 restore_nameidata();
3562 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3563 const char *name, const struct open_flags *op)
3565 struct nameidata nd;
3567 struct filename *filename;
3568 int flags = op->lookup_flags | LOOKUP_ROOT;
3571 nd.root.dentry = dentry;
3573 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3574 return ERR_PTR(-ELOOP);
3576 filename = getname_kernel(name);
3577 if (IS_ERR(filename))
3578 return ERR_CAST(filename);
3580 set_nameidata(&nd, -1, filename);
3581 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3582 if (unlikely(file == ERR_PTR(-ECHILD)))
3583 file = path_openat(&nd, op, flags);
3584 if (unlikely(file == ERR_PTR(-ESTALE)))
3585 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3586 restore_nameidata();
3591 static struct dentry *filename_create(int dfd, struct filename *name,
3592 struct path *path, unsigned int lookup_flags)
3594 struct dentry *dentry = ERR_PTR(-EEXIST);
3599 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3602 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3603 * other flags passed in are ignored!
3605 lookup_flags &= LOOKUP_REVAL;
3607 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3609 return ERR_CAST(name);
3612 * Yucky last component or no last component at all?
3613 * (foo/., foo/.., /////)
3615 if (unlikely(type != LAST_NORM))
3618 /* don't fail immediately if it's r/o, at least try to report other errors */
3619 err2 = mnt_want_write(path->mnt);
3621 * Do the final lookup.
3623 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3624 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3625 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3630 if (d_is_positive(dentry))
3634 * Special case - lookup gave negative, but... we had foo/bar/
3635 * From the vfs_mknod() POV we just have a negative dentry -
3636 * all is fine. Let's be bastards - you had / on the end, you've
3637 * been asking for (non-existent) directory. -ENOENT for you.
3639 if (unlikely(!is_dir && last.name[last.len])) {
3643 if (unlikely(err2)) {
3651 dentry = ERR_PTR(error);
3653 inode_unlock(path->dentry->d_inode);
3655 mnt_drop_write(path->mnt);
3662 struct dentry *kern_path_create(int dfd, const char *pathname,
3663 struct path *path, unsigned int lookup_flags)
3665 return filename_create(dfd, getname_kernel(pathname),
3666 path, lookup_flags);
3668 EXPORT_SYMBOL(kern_path_create);
3670 void done_path_create(struct path *path, struct dentry *dentry)
3673 inode_unlock(path->dentry->d_inode);
3674 mnt_drop_write(path->mnt);
3677 EXPORT_SYMBOL(done_path_create);
3679 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3680 struct path *path, unsigned int lookup_flags)
3682 return filename_create(dfd, getname(pathname), path, lookup_flags);
3684 EXPORT_SYMBOL(user_path_create);
3686 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3688 int error = may_create(dir, dentry);
3693 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3696 if (!dir->i_op->mknod)
3699 error = devcgroup_inode_mknod(mode, dev);
3703 error = security_inode_mknod(dir, dentry, mode, dev);
3707 error = dir->i_op->mknod(dir, dentry, mode, dev);
3709 fsnotify_create(dir, dentry);
3712 EXPORT_SYMBOL(vfs_mknod);
3714 static int may_mknod(umode_t mode)
3716 switch (mode & S_IFMT) {
3722 case 0: /* zero mode translates to S_IFREG */
3731 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3734 struct dentry *dentry;
3737 unsigned int lookup_flags = 0;
3739 error = may_mknod(mode);
3743 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3745 return PTR_ERR(dentry);
3747 if (!IS_POSIXACL(path.dentry->d_inode))
3748 mode &= ~current_umask();
3749 error = security_path_mknod(&path, dentry, mode, dev);
3752 switch (mode & S_IFMT) {
3753 case 0: case S_IFREG:
3754 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3756 ima_post_path_mknod(dentry);
3758 case S_IFCHR: case S_IFBLK:
3759 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3760 new_decode_dev(dev));
3762 case S_IFIFO: case S_IFSOCK:
3763 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3767 done_path_create(&path, dentry);
3768 if (retry_estale(error, lookup_flags)) {
3769 lookup_flags |= LOOKUP_REVAL;
3775 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3777 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3780 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3782 int error = may_create(dir, dentry);
3783 unsigned max_links = dir->i_sb->s_max_links;
3788 if (!dir->i_op->mkdir)
3791 mode &= (S_IRWXUGO|S_ISVTX);
3792 error = security_inode_mkdir(dir, dentry, mode);
3796 if (max_links && dir->i_nlink >= max_links)
3799 error = dir->i_op->mkdir(dir, dentry, mode);
3801 fsnotify_mkdir(dir, dentry);
3804 EXPORT_SYMBOL(vfs_mkdir);
3806 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3808 struct dentry *dentry;
3811 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3814 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3816 return PTR_ERR(dentry);
3818 if (!IS_POSIXACL(path.dentry->d_inode))
3819 mode &= ~current_umask();
3820 error = security_path_mkdir(&path, dentry, mode);
3822 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3823 done_path_create(&path, dentry);
3824 if (retry_estale(error, lookup_flags)) {
3825 lookup_flags |= LOOKUP_REVAL;
3831 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3833 return sys_mkdirat(AT_FDCWD, pathname, mode);
3836 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3838 int error = may_delete(dir, dentry, 1);
3843 if (!dir->i_op->rmdir)
3847 inode_lock(dentry->d_inode);
3850 if (is_local_mountpoint(dentry))
3853 error = security_inode_rmdir(dir, dentry);
3857 shrink_dcache_parent(dentry);
3858 error = dir->i_op->rmdir(dir, dentry);
3862 dentry->d_inode->i_flags |= S_DEAD;
3864 detach_mounts(dentry);
3867 inode_unlock(dentry->d_inode);
3873 EXPORT_SYMBOL(vfs_rmdir);
3875 static long do_rmdir(int dfd, const char __user *pathname)
3878 struct filename *name;
3879 struct dentry *dentry;
3883 unsigned int lookup_flags = 0;
3885 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3886 &path, &last, &type);
3888 return PTR_ERR(name);
3902 error = mnt_want_write(path.mnt);
3906 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3907 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3908 error = PTR_ERR(dentry);
3911 if (!dentry->d_inode) {
3915 error = security_path_rmdir(&path, dentry);
3918 error = vfs_rmdir(path.dentry->d_inode, dentry);
3922 inode_unlock(path.dentry->d_inode);
3923 mnt_drop_write(path.mnt);
3927 if (retry_estale(error, lookup_flags)) {
3928 lookup_flags |= LOOKUP_REVAL;
3934 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3936 return do_rmdir(AT_FDCWD, pathname);
3940 * vfs_unlink - unlink a filesystem object
3941 * @dir: parent directory
3943 * @delegated_inode: returns victim inode, if the inode is delegated.
3945 * The caller must hold dir->i_mutex.
3947 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3948 * return a reference to the inode in delegated_inode. The caller
3949 * should then break the delegation on that inode and retry. Because
3950 * breaking a delegation may take a long time, the caller should drop
3951 * dir->i_mutex before doing so.
3953 * Alternatively, a caller may pass NULL for delegated_inode. This may
3954 * be appropriate for callers that expect the underlying filesystem not
3955 * to be NFS exported.
3957 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3959 struct inode *target = dentry->d_inode;
3960 int error = may_delete(dir, dentry, 0);
3965 if (!dir->i_op->unlink)
3969 if (is_local_mountpoint(dentry))
3972 error = security_inode_unlink(dir, dentry);
3974 error = try_break_deleg(target, delegated_inode);
3977 error = dir->i_op->unlink(dir, dentry);
3980 detach_mounts(dentry);
3985 inode_unlock(target);
3987 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3988 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3989 fsnotify_link_count(target);
3995 EXPORT_SYMBOL(vfs_unlink);
3998 * Make sure that the actual truncation of the file will occur outside its
3999 * directory's i_mutex. Truncate can take a long time if there is a lot of
4000 * writeout happening, and we don't want to prevent access to the directory
4001 * while waiting on the I/O.
4003 static long do_unlinkat(int dfd, const char __user *pathname)
4006 struct filename *name;
4007 struct dentry *dentry;
4011 struct inode *inode = NULL;
4012 struct inode *delegated_inode = NULL;
4013 unsigned int lookup_flags = 0;
4015 name = filename_parentat(dfd, getname(pathname), lookup_flags,
4016 &path, &last, &type);
4018 return PTR_ERR(name);
4021 if (type != LAST_NORM)
4024 error = mnt_want_write(path.mnt);
4028 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4029 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4030 error = PTR_ERR(dentry);
4031 if (!IS_ERR(dentry)) {
4032 /* Why not before? Because we want correct error value */
4033 if (last.name[last.len])
4035 inode = dentry->d_inode;
4036 if (d_is_negative(dentry))
4039 error = security_path_unlink(&path, dentry);
4042 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4046 inode_unlock(path.dentry->d_inode);
4048 iput(inode); /* truncate the inode here */
4050 if (delegated_inode) {
4051 error = break_deleg_wait(&delegated_inode);
4055 mnt_drop_write(path.mnt);
4059 if (retry_estale(error, lookup_flags)) {
4060 lookup_flags |= LOOKUP_REVAL;
4067 if (d_is_negative(dentry))
4069 else if (d_is_dir(dentry))
4076 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4078 if ((flag & ~AT_REMOVEDIR) != 0)
4081 if (flag & AT_REMOVEDIR)
4082 return do_rmdir(dfd, pathname);
4084 return do_unlinkat(dfd, pathname);
4087 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4089 return do_unlinkat(AT_FDCWD, pathname);
4092 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4094 int error = may_create(dir, dentry);
4099 if (!dir->i_op->symlink)
4102 error = security_inode_symlink(dir, dentry, oldname);
4106 error = dir->i_op->symlink(dir, dentry, oldname);
4108 fsnotify_create(dir, dentry);
4111 EXPORT_SYMBOL(vfs_symlink);
4113 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4114 int, newdfd, const char __user *, newname)
4117 struct filename *from;
4118 struct dentry *dentry;
4120 unsigned int lookup_flags = 0;
4122 from = getname(oldname);
4124 return PTR_ERR(from);
4126 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4127 error = PTR_ERR(dentry);
4131 error = security_path_symlink(&path, dentry, from->name);
4133 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4134 done_path_create(&path, dentry);
4135 if (retry_estale(error, lookup_flags)) {
4136 lookup_flags |= LOOKUP_REVAL;
4144 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4146 return sys_symlinkat(oldname, AT_FDCWD, newname);
4150 * vfs_link - create a new link
4151 * @old_dentry: object to be linked
4153 * @new_dentry: where to create the new link
4154 * @delegated_inode: returns inode needing a delegation break
4156 * The caller must hold dir->i_mutex
4158 * If vfs_link discovers a delegation on the to-be-linked file in need
4159 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4160 * inode in delegated_inode. The caller should then break the delegation
4161 * and retry. Because breaking a delegation may take a long time, the
4162 * caller should drop the i_mutex before doing so.
4164 * Alternatively, a caller may pass NULL for delegated_inode. This may
4165 * be appropriate for callers that expect the underlying filesystem not
4166 * to be NFS exported.
4168 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4170 struct inode *inode = old_dentry->d_inode;
4171 unsigned max_links = dir->i_sb->s_max_links;
4177 error = may_create(dir, new_dentry);
4181 if (dir->i_sb != inode->i_sb)
4185 * A link to an append-only or immutable file cannot be created.
4187 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4190 * Updating the link count will likely cause i_uid and i_gid to
4191 * be writen back improperly if their true value is unknown to
4194 if (HAS_UNMAPPED_ID(inode))
4196 if (!dir->i_op->link)
4198 if (S_ISDIR(inode->i_mode))
4201 error = security_inode_link(old_dentry, dir, new_dentry);
4206 /* Make sure we don't allow creating hardlink to an unlinked file */
4207 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4209 else if (max_links && inode->i_nlink >= max_links)
4212 error = try_break_deleg(inode, delegated_inode);
4214 error = dir->i_op->link(old_dentry, dir, new_dentry);
4217 if (!error && (inode->i_state & I_LINKABLE)) {
4218 spin_lock(&inode->i_lock);
4219 inode->i_state &= ~I_LINKABLE;
4220 spin_unlock(&inode->i_lock);
4222 inode_unlock(inode);
4224 fsnotify_link(dir, inode, new_dentry);
4227 EXPORT_SYMBOL(vfs_link);
4230 * Hardlinks are often used in delicate situations. We avoid
4231 * security-related surprises by not following symlinks on the
4234 * We don't follow them on the oldname either to be compatible
4235 * with linux 2.0, and to avoid hard-linking to directories
4236 * and other special files. --ADM
4238 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4239 int, newdfd, const char __user *, newname, int, flags)
4241 struct dentry *new_dentry;
4242 struct path old_path, new_path;
4243 struct inode *delegated_inode = NULL;
4247 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4250 * To use null names we require CAP_DAC_READ_SEARCH
4251 * This ensures that not everyone will be able to create
4252 * handlink using the passed filedescriptor.
4254 if (flags & AT_EMPTY_PATH) {
4255 if (!capable(CAP_DAC_READ_SEARCH))
4260 if (flags & AT_SYMLINK_FOLLOW)
4261 how |= LOOKUP_FOLLOW;
4263 error = user_path_at(olddfd, oldname, how, &old_path);
4267 new_dentry = user_path_create(newdfd, newname, &new_path,
4268 (how & LOOKUP_REVAL));
4269 error = PTR_ERR(new_dentry);
4270 if (IS_ERR(new_dentry))
4274 if (old_path.mnt != new_path.mnt)
4276 error = may_linkat(&old_path);
4277 if (unlikely(error))
4279 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4282 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4284 done_path_create(&new_path, new_dentry);
4285 if (delegated_inode) {
4286 error = break_deleg_wait(&delegated_inode);
4288 path_put(&old_path);
4292 if (retry_estale(error, how)) {
4293 path_put(&old_path);
4294 how |= LOOKUP_REVAL;
4298 path_put(&old_path);
4303 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4305 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4309 * vfs_rename - rename a filesystem object
4310 * @old_dir: parent of source
4311 * @old_dentry: source
4312 * @new_dir: parent of destination
4313 * @new_dentry: destination
4314 * @delegated_inode: returns an inode needing a delegation break
4315 * @flags: rename flags
4317 * The caller must hold multiple mutexes--see lock_rename()).
4319 * If vfs_rename discovers a delegation in need of breaking at either
4320 * the source or destination, it will return -EWOULDBLOCK and return a
4321 * reference to the inode in delegated_inode. The caller should then
4322 * break the delegation and retry. Because breaking a delegation may
4323 * take a long time, the caller should drop all locks before doing
4326 * Alternatively, a caller may pass NULL for delegated_inode. This may
4327 * be appropriate for callers that expect the underlying filesystem not
4328 * to be NFS exported.
4330 * The worst of all namespace operations - renaming directory. "Perverted"
4331 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4334 * a) we can get into loop creation.
4335 * b) race potential - two innocent renames can create a loop together.
4336 * That's where 4.4 screws up. Current fix: serialization on
4337 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4339 * c) we have to lock _four_ objects - parents and victim (if it exists),
4340 * and source (if it is not a directory).
4341 * And that - after we got ->i_mutex on parents (until then we don't know
4342 * whether the target exists). Solution: try to be smart with locking
4343 * order for inodes. We rely on the fact that tree topology may change
4344 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4345 * move will be locked. Thus we can rank directories by the tree
4346 * (ancestors first) and rank all non-directories after them.
4347 * That works since everybody except rename does "lock parent, lookup,
4348 * lock child" and rename is under ->s_vfs_rename_mutex.
4349 * HOWEVER, it relies on the assumption that any object with ->lookup()
4350 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4351 * we'd better make sure that there's no link(2) for them.
4352 * d) conversion from fhandle to dentry may come in the wrong moment - when
4353 * we are removing the target. Solution: we will have to grab ->i_mutex
4354 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4355 * ->i_mutex on parents, which works but leads to some truly excessive
4358 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4359 struct inode *new_dir, struct dentry *new_dentry,
4360 struct inode **delegated_inode, unsigned int flags)
4363 bool is_dir = d_is_dir(old_dentry);
4364 struct inode *source = old_dentry->d_inode;
4365 struct inode *target = new_dentry->d_inode;
4366 bool new_is_dir = false;
4367 unsigned max_links = new_dir->i_sb->s_max_links;
4368 struct name_snapshot old_name;
4370 if (source == target)
4373 error = may_delete(old_dir, old_dentry, is_dir);
4378 error = may_create(new_dir, new_dentry);
4380 new_is_dir = d_is_dir(new_dentry);
4382 if (!(flags & RENAME_EXCHANGE))
4383 error = may_delete(new_dir, new_dentry, is_dir);
4385 error = may_delete(new_dir, new_dentry, new_is_dir);
4390 if (!old_dir->i_op->rename)
4394 * If we are going to change the parent - check write permissions,
4395 * we'll need to flip '..'.
4397 if (new_dir != old_dir) {
4399 error = inode_permission(source, MAY_WRITE);
4403 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4404 error = inode_permission(target, MAY_WRITE);
4410 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4415 take_dentry_name_snapshot(&old_name, old_dentry);
4417 if (!is_dir || (flags & RENAME_EXCHANGE))
4418 lock_two_nondirectories(source, target);
4423 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4426 if (max_links && new_dir != old_dir) {
4428 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4430 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4431 old_dir->i_nlink >= max_links)
4434 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4435 shrink_dcache_parent(new_dentry);
4437 error = try_break_deleg(source, delegated_inode);
4441 if (target && !new_is_dir) {
4442 error = try_break_deleg(target, delegated_inode);
4446 error = old_dir->i_op->rename(old_dir, old_dentry,
4447 new_dir, new_dentry, flags);
4451 if (!(flags & RENAME_EXCHANGE) && target) {
4453 target->i_flags |= S_DEAD;
4454 dont_mount(new_dentry);
4455 detach_mounts(new_dentry);
4457 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4458 if (!(flags & RENAME_EXCHANGE))
4459 d_move(old_dentry, new_dentry);
4461 d_exchange(old_dentry, new_dentry);
4464 if (!is_dir || (flags & RENAME_EXCHANGE))
4465 unlock_two_nondirectories(source, target);
4467 inode_unlock(target);
4470 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4471 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4472 if (flags & RENAME_EXCHANGE) {
4473 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4474 new_is_dir, NULL, new_dentry);
4477 release_dentry_name_snapshot(&old_name);
4481 EXPORT_SYMBOL(vfs_rename);
4483 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4484 int, newdfd, const char __user *, newname, unsigned int, flags)
4486 struct dentry *old_dentry, *new_dentry;
4487 struct dentry *trap;
4488 struct path old_path, new_path;
4489 struct qstr old_last, new_last;
4490 int old_type, new_type;
4491 struct inode *delegated_inode = NULL;
4492 struct filename *from;
4493 struct filename *to;
4494 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4495 bool should_retry = false;
4498 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4501 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4502 (flags & RENAME_EXCHANGE))
4505 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4508 if (flags & RENAME_EXCHANGE)
4512 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4513 &old_path, &old_last, &old_type);
4515 error = PTR_ERR(from);
4519 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4520 &new_path, &new_last, &new_type);
4522 error = PTR_ERR(to);
4527 if (old_path.mnt != new_path.mnt)
4531 if (old_type != LAST_NORM)
4534 if (flags & RENAME_NOREPLACE)
4536 if (new_type != LAST_NORM)
4539 error = mnt_want_write(old_path.mnt);
4544 trap = lock_rename(new_path.dentry, old_path.dentry);
4546 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4547 error = PTR_ERR(old_dentry);
4548 if (IS_ERR(old_dentry))
4550 /* source must exist */
4552 if (d_is_negative(old_dentry))
4554 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4555 error = PTR_ERR(new_dentry);
4556 if (IS_ERR(new_dentry))
4559 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4561 if (flags & RENAME_EXCHANGE) {
4563 if (d_is_negative(new_dentry))
4566 if (!d_is_dir(new_dentry)) {
4568 if (new_last.name[new_last.len])
4572 /* unless the source is a directory trailing slashes give -ENOTDIR */
4573 if (!d_is_dir(old_dentry)) {
4575 if (old_last.name[old_last.len])
4577 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4580 /* source should not be ancestor of target */
4582 if (old_dentry == trap)
4584 /* target should not be an ancestor of source */
4585 if (!(flags & RENAME_EXCHANGE))
4587 if (new_dentry == trap)
4590 error = security_path_rename(&old_path, old_dentry,
4591 &new_path, new_dentry, flags);
4594 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4595 new_path.dentry->d_inode, new_dentry,
4596 &delegated_inode, flags);
4602 unlock_rename(new_path.dentry, old_path.dentry);
4603 if (delegated_inode) {
4604 error = break_deleg_wait(&delegated_inode);
4608 mnt_drop_write(old_path.mnt);
4610 if (retry_estale(error, lookup_flags))
4611 should_retry = true;
4612 path_put(&new_path);
4615 path_put(&old_path);
4618 should_retry = false;
4619 lookup_flags |= LOOKUP_REVAL;
4626 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4627 int, newdfd, const char __user *, newname)
4629 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4632 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4634 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4637 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4639 int error = may_create(dir, dentry);
4643 if (!dir->i_op->mknod)
4646 return dir->i_op->mknod(dir, dentry,
4647 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4649 EXPORT_SYMBOL(vfs_whiteout);
4651 int readlink_copy(char __user *buffer, int buflen, const char *link)
4653 int len = PTR_ERR(link);
4658 if (len > (unsigned) buflen)
4660 if (copy_to_user(buffer, link, len))
4667 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4668 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4669 * for any given inode is up to filesystem.
4671 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4674 DEFINE_DELAYED_CALL(done);
4675 struct inode *inode = d_inode(dentry);
4676 const char *link = inode->i_link;
4680 link = inode->i_op->get_link(dentry, inode, &done);
4682 return PTR_ERR(link);
4684 res = readlink_copy(buffer, buflen, link);
4685 do_delayed_call(&done);
4690 * vfs_readlink - copy symlink body into userspace buffer
4691 * @dentry: dentry on which to get symbolic link
4692 * @buffer: user memory pointer
4693 * @buflen: size of buffer
4695 * Does not touch atime. That's up to the caller if necessary
4697 * Does not call security hook.
4699 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4701 struct inode *inode = d_inode(dentry);
4703 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4704 if (unlikely(inode->i_op->readlink))
4705 return inode->i_op->readlink(dentry, buffer, buflen);
4707 if (!d_is_symlink(dentry))
4710 spin_lock(&inode->i_lock);
4711 inode->i_opflags |= IOP_DEFAULT_READLINK;
4712 spin_unlock(&inode->i_lock);
4715 return generic_readlink(dentry, buffer, buflen);
4717 EXPORT_SYMBOL(vfs_readlink);
4720 * vfs_get_link - get symlink body
4721 * @dentry: dentry on which to get symbolic link
4722 * @done: caller needs to free returned data with this
4724 * Calls security hook and i_op->get_link() on the supplied inode.
4726 * It does not touch atime. That's up to the caller if necessary.
4728 * Does not work on "special" symlinks like /proc/$$/fd/N
4730 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4732 const char *res = ERR_PTR(-EINVAL);
4733 struct inode *inode = d_inode(dentry);
4735 if (d_is_symlink(dentry)) {
4736 res = ERR_PTR(security_inode_readlink(dentry));
4738 res = inode->i_op->get_link(dentry, inode, done);
4742 EXPORT_SYMBOL(vfs_get_link);
4744 /* get the link contents into pagecache */
4745 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4746 struct delayed_call *callback)
4750 struct address_space *mapping = inode->i_mapping;
4753 page = find_get_page(mapping, 0);
4755 return ERR_PTR(-ECHILD);
4756 if (!PageUptodate(page)) {
4758 return ERR_PTR(-ECHILD);
4761 page = read_mapping_page(mapping, 0, NULL);
4765 set_delayed_call(callback, page_put_link, page);
4766 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4767 kaddr = page_address(page);
4768 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4772 EXPORT_SYMBOL(page_get_link);
4774 void page_put_link(void *arg)
4778 EXPORT_SYMBOL(page_put_link);
4780 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4782 DEFINE_DELAYED_CALL(done);
4783 int res = readlink_copy(buffer, buflen,
4784 page_get_link(dentry, d_inode(dentry),
4786 do_delayed_call(&done);
4789 EXPORT_SYMBOL(page_readlink);
4792 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4794 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4796 struct address_space *mapping = inode->i_mapping;
4800 unsigned int flags = 0;
4802 flags |= AOP_FLAG_NOFS;
4805 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4806 flags, &page, &fsdata);
4810 memcpy(page_address(page), symname, len-1);
4812 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4819 mark_inode_dirty(inode);
4824 EXPORT_SYMBOL(__page_symlink);
4826 int page_symlink(struct inode *inode, const char *symname, int len)
4828 return __page_symlink(inode, symname, len,
4829 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4831 EXPORT_SYMBOL(page_symlink);
4833 const struct inode_operations page_symlink_inode_operations = {
4834 .get_link = page_get_link,
4836 EXPORT_SYMBOL(page_symlink_inode_operations);