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 * sb_permission - Check superblock-level permissions
396 * @sb: Superblock of inode to check permission on
397 * @inode: Inode to check permission on
398 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
400 * Separate out file-system wide checks from inode-specific permission checks.
402 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
404 if (unlikely(mask & MAY_WRITE)) {
405 umode_t mode = inode->i_mode;
407 /* Nobody gets write access to a read-only fs. */
408 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
415 * inode_permission - Check for access rights to a given inode
416 * @inode: Inode to check permission on
417 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
419 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
420 * this, letting us set arbitrary permissions for filesystem access without
421 * changing the "normal" UIDs which are used for other things.
423 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
425 int inode_permission(struct inode *inode, int mask)
429 retval = sb_permission(inode->i_sb, inode, mask);
433 if (unlikely(mask & MAY_WRITE)) {
435 * Nobody gets write access to an immutable file.
437 if (IS_IMMUTABLE(inode))
441 * Updating mtime will likely cause i_uid and i_gid to be
442 * written back improperly if their true value is unknown
445 if (HAS_UNMAPPED_ID(inode))
449 retval = do_inode_permission(inode, mask);
453 retval = devcgroup_inode_permission(inode, mask);
457 return security_inode_permission(inode, mask);
459 EXPORT_SYMBOL(inode_permission);
462 * path_get - get a reference to a path
463 * @path: path to get the reference to
465 * Given a path increment the reference count to the dentry and the vfsmount.
467 void path_get(const struct path *path)
472 EXPORT_SYMBOL(path_get);
475 * path_put - put a reference to a path
476 * @path: path to put the reference to
478 * Given a path decrement the reference count to the dentry and the vfsmount.
480 void path_put(const struct path *path)
485 EXPORT_SYMBOL(path_put);
487 #define EMBEDDED_LEVELS 2
492 struct inode *inode; /* path.dentry.d_inode */
497 int total_link_count;
500 struct delayed_call done;
503 } *stack, internal[EMBEDDED_LEVELS];
504 struct filename *name;
505 struct nameidata *saved;
506 struct inode *link_inode;
509 } __randomize_layout;
511 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
513 struct nameidata *old = current->nameidata;
514 p->stack = p->internal;
517 p->total_link_count = old ? old->total_link_count : 0;
519 current->nameidata = p;
522 static void restore_nameidata(void)
524 struct nameidata *now = current->nameidata, *old = now->saved;
526 current->nameidata = old;
528 old->total_link_count = now->total_link_count;
529 if (now->stack != now->internal)
533 static int __nd_alloc_stack(struct nameidata *nd)
537 if (nd->flags & LOOKUP_RCU) {
538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
543 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
548 memcpy(p, nd->internal, sizeof(nd->internal));
554 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
555 * @path: nameidate to verify
557 * Rename can sometimes move a file or directory outside of a bind
558 * mount, path_connected allows those cases to be detected.
560 static bool path_connected(const struct path *path)
562 struct vfsmount *mnt = path->mnt;
563 struct super_block *sb = mnt->mnt_sb;
565 /* Bind mounts and multi-root filesystems can have disconnected paths */
566 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
569 return is_subdir(path->dentry, mnt->mnt_root);
572 static inline int nd_alloc_stack(struct nameidata *nd)
574 if (likely(nd->depth != EMBEDDED_LEVELS))
576 if (likely(nd->stack != nd->internal))
578 return __nd_alloc_stack(nd);
581 static void drop_links(struct nameidata *nd)
585 struct saved *last = nd->stack + i;
586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
591 static void terminate_walk(struct nameidata *nd)
594 if (!(nd->flags & LOOKUP_RCU)) {
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
604 nd->flags &= ~LOOKUP_RCU;
610 /* path_put is needed afterwards regardless of success or failure */
611 static bool legitimize_path(struct nameidata *nd,
612 struct path *path, unsigned seq)
614 int res = __legitimize_mnt(path->mnt, nd->m_seq);
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
628 static bool legitimize_links(struct nameidata *nd)
631 for (i = 0; i < nd->depth; i++) {
632 struct saved *last = nd->stack + i;
633 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
642 static bool legitimize_root(struct nameidata *nd)
644 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
646 nd->flags |= LOOKUP_ROOT_GRABBED;
647 return legitimize_path(nd, &nd->root, nd->root_seq);
651 * Path walking has 2 modes, rcu-walk and ref-walk (see
652 * Documentation/filesystems/path-lookup.txt). In situations when we can't
653 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
654 * normal reference counts on dentries and vfsmounts to transition to ref-walk
655 * mode. Refcounts are grabbed at the last known good point before rcu-walk
656 * got stuck, so ref-walk may continue from there. If this is not successful
657 * (eg. a seqcount has changed), then failure is returned and it's up to caller
658 * to restart the path walk from the beginning in ref-walk mode.
662 * unlazy_walk - try to switch to ref-walk mode.
663 * @nd: nameidata pathwalk data
664 * Returns: 0 on success, -ECHILD on failure
666 * unlazy_walk attempts to legitimize the current nd->path and nd->root
668 * Must be called from rcu-walk context.
669 * Nothing should touch nameidata between unlazy_walk() failure and
672 static int unlazy_walk(struct nameidata *nd)
674 struct dentry *parent = nd->path.dentry;
676 BUG_ON(!(nd->flags & LOOKUP_RCU));
678 nd->flags &= ~LOOKUP_RCU;
679 if (unlikely(!legitimize_links(nd)))
681 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
683 if (unlikely(!legitimize_root(nd)))
686 BUG_ON(nd->inode != parent->d_inode);
691 nd->path.dentry = NULL;
698 * unlazy_child - try to switch to ref-walk mode.
699 * @nd: nameidata pathwalk data
700 * @dentry: child of nd->path.dentry
701 * @seq: seq number to check dentry against
702 * Returns: 0 on success, -ECHILD on failure
704 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
705 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
706 * @nd. Must be called from rcu-walk context.
707 * Nothing should touch nameidata between unlazy_child() failure and
710 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
712 BUG_ON(!(nd->flags & LOOKUP_RCU));
714 nd->flags &= ~LOOKUP_RCU;
715 if (unlikely(!legitimize_links(nd)))
717 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
719 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
723 * We need to move both the parent and the dentry from the RCU domain
724 * to be properly refcounted. And the sequence number in the dentry
725 * validates *both* dentry counters, since we checked the sequence
726 * number of the parent after we got the child sequence number. So we
727 * know the parent must still be valid if the child sequence number is
729 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
731 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
734 * Sequence counts matched. Now make sure that the root is
735 * still valid and get it if required.
737 if (unlikely(!legitimize_root(nd)))
745 nd->path.dentry = NULL;
755 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
757 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
758 return dentry->d_op->d_revalidate(dentry, flags);
764 * complete_walk - successful completion of path walk
765 * @nd: pointer nameidata
767 * If we had been in RCU mode, drop out of it and legitimize nd->path.
768 * Revalidate the final result, unless we'd already done that during
769 * the path walk or the filesystem doesn't ask for it. Return 0 on
770 * success, -error on failure. In case of failure caller does not
771 * need to drop nd->path.
773 static int complete_walk(struct nameidata *nd)
775 struct dentry *dentry = nd->path.dentry;
778 if (nd->flags & LOOKUP_RCU) {
779 if (!(nd->flags & LOOKUP_ROOT))
781 if (unlikely(unlazy_walk(nd)))
785 if (likely(!(nd->flags & LOOKUP_JUMPED)))
788 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
791 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
801 static void set_root(struct nameidata *nd)
803 struct fs_struct *fs = current->fs;
805 if (nd->flags & LOOKUP_RCU) {
809 seq = read_seqcount_begin(&fs->seq);
811 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
812 } while (read_seqcount_retry(&fs->seq, seq));
814 get_fs_root(fs, &nd->root);
815 nd->flags |= LOOKUP_ROOT_GRABBED;
819 static void path_put_conditional(struct path *path, struct nameidata *nd)
822 if (path->mnt != nd->path.mnt)
826 static inline void path_to_nameidata(const struct path *path,
827 struct nameidata *nd)
829 if (!(nd->flags & LOOKUP_RCU)) {
830 dput(nd->path.dentry);
831 if (nd->path.mnt != path->mnt)
832 mntput(nd->path.mnt);
834 nd->path.mnt = path->mnt;
835 nd->path.dentry = path->dentry;
838 static int nd_jump_root(struct nameidata *nd)
840 if (nd->flags & LOOKUP_RCU) {
844 nd->inode = d->d_inode;
845 nd->seq = nd->root_seq;
846 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
852 nd->inode = nd->path.dentry->d_inode;
854 nd->flags |= LOOKUP_JUMPED;
859 * Helper to directly jump to a known parsed path from ->get_link,
860 * caller must have taken a reference to path beforehand.
862 void nd_jump_link(struct path *path)
864 struct nameidata *nd = current->nameidata;
868 nd->inode = nd->path.dentry->d_inode;
869 nd->flags |= LOOKUP_JUMPED;
872 static inline void put_link(struct nameidata *nd)
874 struct saved *last = nd->stack + --nd->depth;
875 do_delayed_call(&last->done);
876 if (!(nd->flags & LOOKUP_RCU))
877 path_put(&last->link);
880 int sysctl_protected_symlinks __read_mostly = 0;
881 int sysctl_protected_hardlinks __read_mostly = 0;
882 int sysctl_protected_fifos __read_mostly;
883 int sysctl_protected_regular __read_mostly;
886 * may_follow_link - Check symlink following for unsafe situations
887 * @nd: nameidata pathwalk data
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
898 * Returns 0 if following the symlink is allowed, -ve on error.
900 static inline int may_follow_link(struct nameidata *nd)
902 const struct inode *inode;
903 const struct inode *parent;
906 if (!sysctl_protected_symlinks)
909 /* Allowed if owner and follower match. */
910 inode = nd->link_inode;
911 if (uid_eq(current_cred()->fsuid, inode->i_uid))
914 /* Allowed if parent directory not sticky and world-writable. */
916 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
919 /* Allowed if parent directory and link owner match. */
920 puid = parent->i_uid;
921 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
924 if (nd->flags & LOOKUP_RCU)
927 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
928 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
933 * safe_hardlink_source - Check for safe hardlink conditions
934 * @inode: the source inode to hardlink from
936 * Return false if at least one of the following conditions:
937 * - inode is not a regular file
939 * - inode is setgid and group-exec
940 * - access failure for read and write
942 * Otherwise returns true.
944 static bool safe_hardlink_source(struct inode *inode)
946 umode_t mode = inode->i_mode;
948 /* Special files should not get pinned to the filesystem. */
952 /* Setuid files should not get pinned to the filesystem. */
956 /* Executable setgid files should not get pinned to the filesystem. */
957 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
960 /* Hardlinking to unreadable or unwritable sources is dangerous. */
961 if (inode_permission(inode, MAY_READ | MAY_WRITE))
968 * may_linkat - Check permissions for creating a hardlink
969 * @link: the source to hardlink from
971 * Block hardlink when all of:
972 * - sysctl_protected_hardlinks enabled
973 * - fsuid does not match inode
974 * - hardlink source is unsafe (see safe_hardlink_source() above)
975 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
977 * Returns 0 if successful, -ve on error.
979 static int may_linkat(struct path *link)
981 struct inode *inode = link->dentry->d_inode;
983 /* Inode writeback is not safe when the uid or gid are invalid. */
984 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
987 if (!sysctl_protected_hardlinks)
990 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
991 * otherwise, it must be a safe source.
993 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
996 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1001 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1002 * should be allowed, or not, on files that already
1004 * @dir: the sticky parent directory
1005 * @inode: the inode of the file to open
1007 * Block an O_CREAT open of a FIFO (or a regular file) when:
1008 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1009 * - the file already exists
1010 * - we are in a sticky directory
1011 * - we don't own the file
1012 * - the owner of the directory doesn't own the file
1013 * - the directory is world writable
1014 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1015 * the directory doesn't have to be world writable: being group writable will
1018 * Returns 0 if the open is allowed, -ve on error.
1020 static int may_create_in_sticky(struct dentry * const dir,
1021 struct inode * const inode)
1023 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1024 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1025 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1026 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1027 uid_eq(current_fsuid(), inode->i_uid))
1030 if (likely(dir->d_inode->i_mode & 0002) ||
1031 (dir->d_inode->i_mode & 0020 &&
1032 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1033 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1034 const char *operation = S_ISFIFO(inode->i_mode) ?
1035 "sticky_create_fifo" :
1036 "sticky_create_regular";
1037 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1043 static __always_inline
1044 const char *get_link(struct nameidata *nd)
1046 struct saved *last = nd->stack + nd->depth - 1;
1047 struct dentry *dentry = last->link.dentry;
1048 struct inode *inode = nd->link_inode;
1052 if (!(nd->flags & LOOKUP_RCU)) {
1053 touch_atime(&last->link);
1055 } else if (atime_needs_update(&last->link, inode)) {
1056 if (unlikely(unlazy_walk(nd)))
1057 return ERR_PTR(-ECHILD);
1058 touch_atime(&last->link);
1061 error = security_inode_follow_link(dentry, inode,
1062 nd->flags & LOOKUP_RCU);
1063 if (unlikely(error))
1064 return ERR_PTR(error);
1066 nd->last_type = LAST_BIND;
1067 res = READ_ONCE(inode->i_link);
1069 const char * (*get)(struct dentry *, struct inode *,
1070 struct delayed_call *);
1071 get = inode->i_op->get_link;
1072 if (nd->flags & LOOKUP_RCU) {
1073 res = get(NULL, inode, &last->done);
1074 if (res == ERR_PTR(-ECHILD)) {
1075 if (unlikely(unlazy_walk(nd)))
1076 return ERR_PTR(-ECHILD);
1077 res = get(dentry, inode, &last->done);
1080 res = get(dentry, inode, &last->done);
1082 if (IS_ERR_OR_NULL(res))
1088 if (unlikely(nd_jump_root(nd)))
1089 return ERR_PTR(-ECHILD);
1090 while (unlikely(*++res == '/'))
1099 * follow_up - Find the mountpoint of path's vfsmount
1101 * Given a path, find the mountpoint of its source file system.
1102 * Replace @path with the path of the mountpoint in the parent mount.
1105 * Return 1 if we went up a level and 0 if we were already at the
1108 int follow_up(struct path *path)
1110 struct mount *mnt = real_mount(path->mnt);
1111 struct mount *parent;
1112 struct dentry *mountpoint;
1114 read_seqlock_excl(&mount_lock);
1115 parent = mnt->mnt_parent;
1116 if (parent == mnt) {
1117 read_sequnlock_excl(&mount_lock);
1120 mntget(&parent->mnt);
1121 mountpoint = dget(mnt->mnt_mountpoint);
1122 read_sequnlock_excl(&mount_lock);
1124 path->dentry = mountpoint;
1126 path->mnt = &parent->mnt;
1129 EXPORT_SYMBOL(follow_up);
1132 * Perform an automount
1133 * - return -EISDIR to tell follow_managed() to stop and return the path we
1136 static int follow_automount(struct path *path, struct nameidata *nd,
1139 struct vfsmount *mnt;
1142 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1145 /* We don't want to mount if someone's just doing a stat -
1146 * unless they're stat'ing a directory and appended a '/' to
1149 * We do, however, want to mount if someone wants to open or
1150 * create a file of any type under the mountpoint, wants to
1151 * traverse through the mountpoint or wants to open the
1152 * mounted directory. Also, autofs may mark negative dentries
1153 * as being automount points. These will need the attentions
1154 * of the daemon to instantiate them before they can be used.
1156 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1157 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1158 path->dentry->d_inode)
1161 nd->total_link_count++;
1162 if (nd->total_link_count >= 40)
1165 mnt = path->dentry->d_op->d_automount(path);
1168 * The filesystem is allowed to return -EISDIR here to indicate
1169 * it doesn't want to automount. For instance, autofs would do
1170 * this so that its userspace daemon can mount on this dentry.
1172 * However, we can only permit this if it's a terminal point in
1173 * the path being looked up; if it wasn't then the remainder of
1174 * the path is inaccessible and we should say so.
1176 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1178 return PTR_ERR(mnt);
1181 if (!mnt) /* mount collision */
1184 if (!*need_mntput) {
1185 /* lock_mount() may release path->mnt on error */
1187 *need_mntput = true;
1189 err = finish_automount(mnt, path);
1193 /* Someone else made a mount here whilst we were busy */
1198 path->dentry = dget(mnt->mnt_root);
1207 * Handle a dentry that is managed in some way.
1208 * - Flagged for transit management (autofs)
1209 * - Flagged as mountpoint
1210 * - Flagged as automount point
1212 * This may only be called in refwalk mode.
1213 * On success path->dentry is known positive.
1215 * Serialization is taken care of in namespace.c
1217 static int follow_managed(struct path *path, struct nameidata *nd)
1219 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1221 bool need_mntput = false;
1224 /* Given that we're not holding a lock here, we retain the value in a
1225 * local variable for each dentry as we look at it so that we don't see
1226 * the components of that value change under us */
1227 while (flags = smp_load_acquire(&path->dentry->d_flags),
1228 unlikely(flags & DCACHE_MANAGED_DENTRY)) {
1229 /* Allow the filesystem to manage the transit without i_mutex
1231 if (flags & DCACHE_MANAGE_TRANSIT) {
1232 BUG_ON(!path->dentry->d_op);
1233 BUG_ON(!path->dentry->d_op->d_manage);
1234 ret = path->dentry->d_op->d_manage(path, false);
1235 flags = smp_load_acquire(&path->dentry->d_flags);
1240 /* Transit to a mounted filesystem. */
1241 if (flags & DCACHE_MOUNTED) {
1242 struct vfsmount *mounted = lookup_mnt(path);
1247 path->mnt = mounted;
1248 path->dentry = dget(mounted->mnt_root);
1253 /* Something is mounted on this dentry in another
1254 * namespace and/or whatever was mounted there in this
1255 * namespace got unmounted before lookup_mnt() could
1259 /* Handle an automount point */
1260 if (flags & DCACHE_NEED_AUTOMOUNT) {
1261 ret = follow_automount(path, nd, &need_mntput);
1267 /* We didn't change the current path point */
1271 if (need_mntput && path->mnt == mnt)
1274 nd->flags |= LOOKUP_JUMPED;
1275 if (ret == -EISDIR || !ret)
1277 if (ret > 0 && unlikely(d_flags_negative(flags)))
1279 if (unlikely(ret < 0))
1280 path_put_conditional(path, nd);
1284 int follow_down_one(struct path *path)
1286 struct vfsmount *mounted;
1288 mounted = lookup_mnt(path);
1292 path->mnt = mounted;
1293 path->dentry = dget(mounted->mnt_root);
1298 EXPORT_SYMBOL(follow_down_one);
1300 static inline int managed_dentry_rcu(const struct path *path)
1302 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1303 path->dentry->d_op->d_manage(path, true) : 0;
1307 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1308 * we meet a managed dentry that would need blocking.
1310 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1311 struct inode **inode, unsigned *seqp)
1314 struct mount *mounted;
1316 * Don't forget we might have a non-mountpoint managed dentry
1317 * that wants to block transit.
1319 switch (managed_dentry_rcu(path)) {
1329 if (!d_mountpoint(path->dentry))
1330 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1332 mounted = __lookup_mnt(path->mnt, path->dentry);
1335 path->mnt = &mounted->mnt;
1336 path->dentry = mounted->mnt.mnt_root;
1337 nd->flags |= LOOKUP_JUMPED;
1338 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1340 * Update the inode too. We don't need to re-check the
1341 * dentry sequence number here after this d_inode read,
1342 * because a mount-point is always pinned.
1344 *inode = path->dentry->d_inode;
1346 return !read_seqretry(&mount_lock, nd->m_seq) &&
1347 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1350 static int follow_dotdot_rcu(struct nameidata *nd)
1352 struct inode *inode = nd->inode;
1355 if (path_equal(&nd->path, &nd->root))
1357 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1358 struct dentry *old = nd->path.dentry;
1359 struct dentry *parent = old->d_parent;
1362 inode = parent->d_inode;
1363 seq = read_seqcount_begin(&parent->d_seq);
1364 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1366 nd->path.dentry = parent;
1368 if (unlikely(!path_connected(&nd->path)))
1372 struct mount *mnt = real_mount(nd->path.mnt);
1373 struct mount *mparent = mnt->mnt_parent;
1374 struct dentry *mountpoint = mnt->mnt_mountpoint;
1375 struct inode *inode2 = mountpoint->d_inode;
1376 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1377 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1379 if (&mparent->mnt == nd->path.mnt)
1381 /* we know that mountpoint was pinned */
1382 nd->path.dentry = mountpoint;
1383 nd->path.mnt = &mparent->mnt;
1388 while (unlikely(d_mountpoint(nd->path.dentry))) {
1389 struct mount *mounted;
1390 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1391 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1395 nd->path.mnt = &mounted->mnt;
1396 nd->path.dentry = mounted->mnt.mnt_root;
1397 inode = nd->path.dentry->d_inode;
1398 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1405 * Follow down to the covering mount currently visible to userspace. At each
1406 * point, the filesystem owning that dentry may be queried as to whether the
1407 * caller is permitted to proceed or not.
1409 int follow_down(struct path *path)
1414 while (managed = READ_ONCE(path->dentry->d_flags),
1415 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1416 /* Allow the filesystem to manage the transit without i_mutex
1419 * We indicate to the filesystem if someone is trying to mount
1420 * something here. This gives autofs the chance to deny anyone
1421 * other than its daemon the right to mount on its
1424 * The filesystem may sleep at this point.
1426 if (managed & DCACHE_MANAGE_TRANSIT) {
1427 BUG_ON(!path->dentry->d_op);
1428 BUG_ON(!path->dentry->d_op->d_manage);
1429 ret = path->dentry->d_op->d_manage(path, false);
1431 return ret == -EISDIR ? 0 : ret;
1434 /* Transit to a mounted filesystem. */
1435 if (managed & DCACHE_MOUNTED) {
1436 struct vfsmount *mounted = lookup_mnt(path);
1441 path->mnt = mounted;
1442 path->dentry = dget(mounted->mnt_root);
1446 /* Don't handle automount points here */
1451 EXPORT_SYMBOL(follow_down);
1454 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1456 static void follow_mount(struct path *path)
1458 while (d_mountpoint(path->dentry)) {
1459 struct vfsmount *mounted = lookup_mnt(path);
1464 path->mnt = mounted;
1465 path->dentry = dget(mounted->mnt_root);
1469 static int path_parent_directory(struct path *path)
1471 struct dentry *old = path->dentry;
1472 /* rare case of legitimate dget_parent()... */
1473 path->dentry = dget_parent(path->dentry);
1475 if (unlikely(!path_connected(path)))
1480 static int follow_dotdot(struct nameidata *nd)
1483 if (path_equal(&nd->path, &nd->root))
1485 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1486 int ret = path_parent_directory(&nd->path);
1491 if (!follow_up(&nd->path))
1494 follow_mount(&nd->path);
1495 nd->inode = nd->path.dentry->d_inode;
1500 * This looks up the name in dcache and possibly revalidates the found dentry.
1501 * NULL is returned if the dentry does not exist in the cache.
1503 static struct dentry *lookup_dcache(const struct qstr *name,
1507 struct dentry *dentry = d_lookup(dir, name);
1509 int error = d_revalidate(dentry, flags);
1510 if (unlikely(error <= 0)) {
1512 d_invalidate(dentry);
1514 return ERR_PTR(error);
1521 * Parent directory has inode locked exclusive. This is one
1522 * and only case when ->lookup() gets called on non in-lookup
1523 * dentries - as the matter of fact, this only gets called
1524 * when directory is guaranteed to have no in-lookup children
1527 static struct dentry *__lookup_hash(const struct qstr *name,
1528 struct dentry *base, unsigned int flags)
1530 struct dentry *dentry = lookup_dcache(name, base, flags);
1532 struct inode *dir = base->d_inode;
1537 /* Don't create child dentry for a dead directory. */
1538 if (unlikely(IS_DEADDIR(dir)))
1539 return ERR_PTR(-ENOENT);
1541 dentry = d_alloc(base, name);
1542 if (unlikely(!dentry))
1543 return ERR_PTR(-ENOMEM);
1545 old = dir->i_op->lookup(dir, dentry, flags);
1546 if (unlikely(old)) {
1553 static int lookup_fast(struct nameidata *nd,
1554 struct path *path, struct inode **inode,
1557 struct vfsmount *mnt = nd->path.mnt;
1558 struct dentry *dentry, *parent = nd->path.dentry;
1563 * Rename seqlock is not required here because in the off chance
1564 * of a false negative due to a concurrent rename, the caller is
1565 * going to fall back to non-racy lookup.
1567 if (nd->flags & LOOKUP_RCU) {
1570 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1571 if (unlikely(!dentry)) {
1572 if (unlazy_walk(nd))
1578 * This sequence count validates that the inode matches
1579 * the dentry name information from lookup.
1581 *inode = d_backing_inode(dentry);
1582 negative = d_is_negative(dentry);
1583 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1587 * This sequence count validates that the parent had no
1588 * changes while we did the lookup of the dentry above.
1590 * The memory barrier in read_seqcount_begin of child is
1591 * enough, we can use __read_seqcount_retry here.
1593 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1597 status = d_revalidate(dentry, nd->flags);
1598 if (likely(status > 0)) {
1600 * Note: do negative dentry check after revalidation in
1601 * case that drops it.
1603 if (unlikely(negative))
1606 path->dentry = dentry;
1607 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1610 if (unlazy_child(nd, dentry, seq))
1612 if (unlikely(status == -ECHILD))
1613 /* we'd been told to redo it in non-rcu mode */
1614 status = d_revalidate(dentry, nd->flags);
1616 dentry = __d_lookup(parent, &nd->last);
1617 if (unlikely(!dentry))
1619 status = d_revalidate(dentry, nd->flags);
1621 if (unlikely(status <= 0)) {
1623 d_invalidate(dentry);
1629 path->dentry = dentry;
1630 err = follow_managed(path, nd);
1631 if (likely(err > 0))
1632 *inode = d_backing_inode(path->dentry);
1636 /* Fast lookup failed, do it the slow way */
1637 static struct dentry *__lookup_slow(const struct qstr *name,
1641 struct dentry *dentry, *old;
1642 struct inode *inode = dir->d_inode;
1643 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1645 /* Don't go there if it's already dead */
1646 if (unlikely(IS_DEADDIR(inode)))
1647 return ERR_PTR(-ENOENT);
1649 dentry = d_alloc_parallel(dir, name, &wq);
1652 if (unlikely(!d_in_lookup(dentry))) {
1653 int error = d_revalidate(dentry, flags);
1654 if (unlikely(error <= 0)) {
1656 d_invalidate(dentry);
1661 dentry = ERR_PTR(error);
1664 old = inode->i_op->lookup(inode, dentry, flags);
1665 d_lookup_done(dentry);
1666 if (unlikely(old)) {
1674 static struct dentry *lookup_slow(const struct qstr *name,
1678 struct inode *inode = dir->d_inode;
1680 inode_lock_shared(inode);
1681 res = __lookup_slow(name, dir, flags);
1682 inode_unlock_shared(inode);
1686 static inline int may_lookup(struct nameidata *nd)
1688 if (nd->flags & LOOKUP_RCU) {
1689 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1692 if (unlazy_walk(nd))
1695 return inode_permission(nd->inode, MAY_EXEC);
1698 static inline int handle_dots(struct nameidata *nd, int type)
1700 if (type == LAST_DOTDOT) {
1703 if (nd->flags & LOOKUP_RCU) {
1704 return follow_dotdot_rcu(nd);
1706 return follow_dotdot(nd);
1711 static int pick_link(struct nameidata *nd, struct path *link,
1712 struct inode *inode, unsigned seq)
1716 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1717 path_to_nameidata(link, nd);
1720 if (!(nd->flags & LOOKUP_RCU)) {
1721 if (link->mnt == nd->path.mnt)
1724 error = nd_alloc_stack(nd);
1725 if (unlikely(error)) {
1726 if (error == -ECHILD) {
1727 if (unlikely(!legitimize_path(nd, link, seq))) {
1730 nd->flags &= ~LOOKUP_RCU;
1731 nd->path.mnt = NULL;
1732 nd->path.dentry = NULL;
1734 } else if (likely(unlazy_walk(nd)) == 0)
1735 error = nd_alloc_stack(nd);
1743 last = nd->stack + nd->depth++;
1745 clear_delayed_call(&last->done);
1746 nd->link_inode = inode;
1751 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1754 * Do we need to follow links? We _really_ want to be able
1755 * to do this check without having to look at inode->i_op,
1756 * so we keep a cache of "no, this doesn't need follow_link"
1757 * for the common case.
1759 static inline int step_into(struct nameidata *nd, struct path *path,
1760 int flags, struct inode *inode, unsigned seq)
1762 if (!(flags & WALK_MORE) && nd->depth)
1764 if (likely(!d_is_symlink(path->dentry)) ||
1765 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1766 /* not a symlink or should not follow */
1767 path_to_nameidata(path, nd);
1772 /* make sure that d_is_symlink above matches inode */
1773 if (nd->flags & LOOKUP_RCU) {
1774 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1777 return pick_link(nd, path, inode, seq);
1780 static int walk_component(struct nameidata *nd, int flags)
1783 struct inode *inode;
1787 * "." and ".." are special - ".." especially so because it has
1788 * to be able to know about the current root directory and
1789 * parent relationships.
1791 if (unlikely(nd->last_type != LAST_NORM)) {
1792 err = handle_dots(nd, nd->last_type);
1793 if (!(flags & WALK_MORE) && nd->depth)
1797 err = lookup_fast(nd, &path, &inode, &seq);
1798 if (unlikely(err <= 0)) {
1801 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1803 if (IS_ERR(path.dentry))
1804 return PTR_ERR(path.dentry);
1806 path.mnt = nd->path.mnt;
1807 err = follow_managed(&path, nd);
1808 if (unlikely(err < 0))
1811 seq = 0; /* we are already out of RCU mode */
1812 inode = d_backing_inode(path.dentry);
1815 return step_into(nd, &path, flags, inode, seq);
1819 * We can do the critical dentry name comparison and hashing
1820 * operations one word at a time, but we are limited to:
1822 * - Architectures with fast unaligned word accesses. We could
1823 * do a "get_unaligned()" if this helps and is sufficiently
1826 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1827 * do not trap on the (extremely unlikely) case of a page
1828 * crossing operation.
1830 * - Furthermore, we need an efficient 64-bit compile for the
1831 * 64-bit case in order to generate the "number of bytes in
1832 * the final mask". Again, that could be replaced with a
1833 * efficient population count instruction or similar.
1835 #ifdef CONFIG_DCACHE_WORD_ACCESS
1837 #include <asm/word-at-a-time.h>
1841 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1843 #elif defined(CONFIG_64BIT)
1845 * Register pressure in the mixing function is an issue, particularly
1846 * on 32-bit x86, but almost any function requires one state value and
1847 * one temporary. Instead, use a function designed for two state values
1848 * and no temporaries.
1850 * This function cannot create a collision in only two iterations, so
1851 * we have two iterations to achieve avalanche. In those two iterations,
1852 * we have six layers of mixing, which is enough to spread one bit's
1853 * influence out to 2^6 = 64 state bits.
1855 * Rotate constants are scored by considering either 64 one-bit input
1856 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1857 * probability of that delta causing a change to each of the 128 output
1858 * bits, using a sample of random initial states.
1860 * The Shannon entropy of the computed probabilities is then summed
1861 * to produce a score. Ideally, any input change has a 50% chance of
1862 * toggling any given output bit.
1864 * Mixing scores (in bits) for (12,45):
1865 * Input delta: 1-bit 2-bit
1866 * 1 round: 713.3 42542.6
1867 * 2 rounds: 2753.7 140389.8
1868 * 3 rounds: 5954.1 233458.2
1869 * 4 rounds: 7862.6 256672.2
1870 * Perfect: 8192 258048
1871 * (64*128) (64*63/2 * 128)
1873 #define HASH_MIX(x, y, a) \
1875 y ^= x, x = rol64(x,12),\
1876 x += y, y = rol64(y,45),\
1880 * Fold two longs into one 32-bit hash value. This must be fast, but
1881 * latency isn't quite as critical, as there is a fair bit of additional
1882 * work done before the hash value is used.
1884 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1886 y ^= x * GOLDEN_RATIO_64;
1887 y *= GOLDEN_RATIO_64;
1891 #else /* 32-bit case */
1894 * Mixing scores (in bits) for (7,20):
1895 * Input delta: 1-bit 2-bit
1896 * 1 round: 330.3 9201.6
1897 * 2 rounds: 1246.4 25475.4
1898 * 3 rounds: 1907.1 31295.1
1899 * 4 rounds: 2042.3 31718.6
1900 * Perfect: 2048 31744
1901 * (32*64) (32*31/2 * 64)
1903 #define HASH_MIX(x, y, a) \
1905 y ^= x, x = rol32(x, 7),\
1906 x += y, y = rol32(y,20),\
1909 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1911 /* Use arch-optimized multiply if one exists */
1912 return __hash_32(y ^ __hash_32(x));
1918 * Return the hash of a string of known length. This is carfully
1919 * designed to match hash_name(), which is the more critical function.
1920 * In particular, we must end by hashing a final word containing 0..7
1921 * payload bytes, to match the way that hash_name() iterates until it
1922 * finds the delimiter after the name.
1924 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1926 unsigned long a, x = 0, y = (unsigned long)salt;
1931 a = load_unaligned_zeropad(name);
1932 if (len < sizeof(unsigned long))
1935 name += sizeof(unsigned long);
1936 len -= sizeof(unsigned long);
1938 x ^= a & bytemask_from_count(len);
1940 return fold_hash(x, y);
1942 EXPORT_SYMBOL(full_name_hash);
1944 /* Return the "hash_len" (hash and length) of a null-terminated string */
1945 u64 hashlen_string(const void *salt, const char *name)
1947 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1948 unsigned long adata, mask, len;
1949 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1956 len += sizeof(unsigned long);
1958 a = load_unaligned_zeropad(name+len);
1959 } while (!has_zero(a, &adata, &constants));
1961 adata = prep_zero_mask(a, adata, &constants);
1962 mask = create_zero_mask(adata);
1963 x ^= a & zero_bytemask(mask);
1965 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1967 EXPORT_SYMBOL(hashlen_string);
1970 * Calculate the length and hash of the path component, and
1971 * return the "hash_len" as the result.
1973 static inline u64 hash_name(const void *salt, const char *name)
1975 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1976 unsigned long adata, bdata, mask, len;
1977 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1984 len += sizeof(unsigned long);
1986 a = load_unaligned_zeropad(name+len);
1987 b = a ^ REPEAT_BYTE('/');
1988 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1990 adata = prep_zero_mask(a, adata, &constants);
1991 bdata = prep_zero_mask(b, bdata, &constants);
1992 mask = create_zero_mask(adata | bdata);
1993 x ^= a & zero_bytemask(mask);
1995 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1998 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2000 /* Return the hash of a string of known length */
2001 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2003 unsigned long hash = init_name_hash(salt);
2005 hash = partial_name_hash((unsigned char)*name++, hash);
2006 return end_name_hash(hash);
2008 EXPORT_SYMBOL(full_name_hash);
2010 /* Return the "hash_len" (hash and length) of a null-terminated string */
2011 u64 hashlen_string(const void *salt, const char *name)
2013 unsigned long hash = init_name_hash(salt);
2014 unsigned long len = 0, c;
2016 c = (unsigned char)*name;
2019 hash = partial_name_hash(c, hash);
2020 c = (unsigned char)name[len];
2022 return hashlen_create(end_name_hash(hash), len);
2024 EXPORT_SYMBOL(hashlen_string);
2027 * We know there's a real path component here of at least
2030 static inline u64 hash_name(const void *salt, const char *name)
2032 unsigned long hash = init_name_hash(salt);
2033 unsigned long len = 0, c;
2035 c = (unsigned char)*name;
2038 hash = partial_name_hash(c, hash);
2039 c = (unsigned char)name[len];
2040 } while (c && c != '/');
2041 return hashlen_create(end_name_hash(hash), len);
2048 * This is the basic name resolution function, turning a pathname into
2049 * the final dentry. We expect 'base' to be positive and a directory.
2051 * Returns 0 and nd will have valid dentry and mnt on success.
2052 * Returns error and drops reference to input namei data on failure.
2054 static int link_path_walk(const char *name, struct nameidata *nd)
2059 return PTR_ERR(name);
2065 /* At this point we know we have a real path component. */
2070 err = may_lookup(nd);
2074 hash_len = hash_name(nd->path.dentry, name);
2077 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2079 if (name[1] == '.') {
2081 nd->flags |= LOOKUP_JUMPED;
2087 if (likely(type == LAST_NORM)) {
2088 struct dentry *parent = nd->path.dentry;
2089 nd->flags &= ~LOOKUP_JUMPED;
2090 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2091 struct qstr this = { { .hash_len = hash_len }, .name = name };
2092 err = parent->d_op->d_hash(parent, &this);
2095 hash_len = this.hash_len;
2100 nd->last.hash_len = hash_len;
2101 nd->last.name = name;
2102 nd->last_type = type;
2104 name += hashlen_len(hash_len);
2108 * If it wasn't NUL, we know it was '/'. Skip that
2109 * slash, and continue until no more slashes.
2113 } while (unlikely(*name == '/'));
2114 if (unlikely(!*name)) {
2116 /* pathname body, done */
2119 name = nd->stack[nd->depth - 1].name;
2120 /* trailing symlink, done */
2123 /* last component of nested symlink */
2124 err = walk_component(nd, WALK_FOLLOW);
2126 /* not the last component */
2127 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2133 const char *s = get_link(nd);
2142 nd->stack[nd->depth - 1].name = name;
2147 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2148 if (nd->flags & LOOKUP_RCU) {
2149 if (unlazy_walk(nd))
2157 /* must be paired with terminate_walk() */
2158 static const char *path_init(struct nameidata *nd, unsigned flags)
2160 const char *s = nd->name->name;
2163 flags &= ~LOOKUP_RCU;
2164 if (flags & LOOKUP_RCU)
2167 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2168 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2170 if (flags & LOOKUP_ROOT) {
2171 struct dentry *root = nd->root.dentry;
2172 struct inode *inode = root->d_inode;
2173 if (*s && unlikely(!d_can_lookup(root)))
2174 return ERR_PTR(-ENOTDIR);
2175 nd->path = nd->root;
2177 if (flags & LOOKUP_RCU) {
2178 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2179 nd->root_seq = nd->seq;
2180 nd->m_seq = read_seqbegin(&mount_lock);
2182 path_get(&nd->path);
2187 nd->root.mnt = NULL;
2188 nd->path.mnt = NULL;
2189 nd->path.dentry = NULL;
2191 nd->m_seq = read_seqbegin(&mount_lock);
2194 if (likely(!nd_jump_root(nd)))
2196 return ERR_PTR(-ECHILD);
2197 } else if (nd->dfd == AT_FDCWD) {
2198 if (flags & LOOKUP_RCU) {
2199 struct fs_struct *fs = current->fs;
2203 seq = read_seqcount_begin(&fs->seq);
2205 nd->inode = nd->path.dentry->d_inode;
2206 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2207 } while (read_seqcount_retry(&fs->seq, seq));
2209 get_fs_pwd(current->fs, &nd->path);
2210 nd->inode = nd->path.dentry->d_inode;
2214 /* Caller must check execute permissions on the starting path component */
2215 struct fd f = fdget_raw(nd->dfd);
2216 struct dentry *dentry;
2219 return ERR_PTR(-EBADF);
2221 dentry = f.file->f_path.dentry;
2223 if (*s && unlikely(!d_can_lookup(dentry))) {
2225 return ERR_PTR(-ENOTDIR);
2228 nd->path = f.file->f_path;
2229 if (flags & LOOKUP_RCU) {
2230 nd->inode = nd->path.dentry->d_inode;
2231 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2233 path_get(&nd->path);
2234 nd->inode = nd->path.dentry->d_inode;
2241 static const char *trailing_symlink(struct nameidata *nd)
2244 int error = may_follow_link(nd);
2245 if (unlikely(error))
2246 return ERR_PTR(error);
2247 nd->flags |= LOOKUP_PARENT;
2248 nd->stack[0].name = NULL;
2253 static inline int lookup_last(struct nameidata *nd)
2255 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2256 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2258 nd->flags &= ~LOOKUP_PARENT;
2259 return walk_component(nd, 0);
2262 static int handle_lookup_down(struct nameidata *nd)
2264 struct path path = nd->path;
2265 struct inode *inode = nd->inode;
2266 unsigned seq = nd->seq;
2269 if (nd->flags & LOOKUP_RCU) {
2271 * don't bother with unlazy_walk on failure - we are
2272 * at the very beginning of walk, so we lose nothing
2273 * if we simply redo everything in non-RCU mode
2275 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2279 err = follow_managed(&path, nd);
2280 if (unlikely(err < 0))
2282 inode = d_backing_inode(path.dentry);
2285 path_to_nameidata(&path, nd);
2291 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2292 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2294 const char *s = path_init(nd, flags);
2297 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2298 err = handle_lookup_down(nd);
2299 if (unlikely(err < 0))
2303 while (!(err = link_path_walk(s, nd))
2304 && ((err = lookup_last(nd)) > 0)) {
2305 s = trailing_symlink(nd);
2308 err = complete_walk(nd);
2310 if (!err && nd->flags & LOOKUP_DIRECTORY)
2311 if (!d_can_lookup(nd->path.dentry))
2315 nd->path.mnt = NULL;
2316 nd->path.dentry = NULL;
2322 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2323 struct path *path, struct path *root)
2326 struct nameidata nd;
2328 return PTR_ERR(name);
2329 if (unlikely(root)) {
2331 flags |= LOOKUP_ROOT;
2333 set_nameidata(&nd, dfd, name);
2334 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2335 if (unlikely(retval == -ECHILD))
2336 retval = path_lookupat(&nd, flags, path);
2337 if (unlikely(retval == -ESTALE))
2338 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2340 if (likely(!retval))
2341 audit_inode(name, path->dentry, 0);
2342 restore_nameidata();
2347 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2348 static int path_parentat(struct nameidata *nd, unsigned flags,
2349 struct path *parent)
2351 const char *s = path_init(nd, flags);
2352 int err = link_path_walk(s, nd);
2354 err = complete_walk(nd);
2357 nd->path.mnt = NULL;
2358 nd->path.dentry = NULL;
2364 static struct filename *filename_parentat(int dfd, struct filename *name,
2365 unsigned int flags, struct path *parent,
2366 struct qstr *last, int *type)
2369 struct nameidata nd;
2373 set_nameidata(&nd, dfd, name);
2374 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2375 if (unlikely(retval == -ECHILD))
2376 retval = path_parentat(&nd, flags, parent);
2377 if (unlikely(retval == -ESTALE))
2378 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2379 if (likely(!retval)) {
2381 *type = nd.last_type;
2382 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2385 name = ERR_PTR(retval);
2387 restore_nameidata();
2391 /* does lookup, returns the object with parent locked */
2392 struct dentry *kern_path_locked(const char *name, struct path *path)
2394 struct filename *filename;
2399 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2401 if (IS_ERR(filename))
2402 return ERR_CAST(filename);
2403 if (unlikely(type != LAST_NORM)) {
2406 return ERR_PTR(-EINVAL);
2408 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2409 d = __lookup_hash(&last, path->dentry, 0);
2411 inode_unlock(path->dentry->d_inode);
2418 int kern_path(const char *name, unsigned int flags, struct path *path)
2420 return filename_lookup(AT_FDCWD, getname_kernel(name),
2423 EXPORT_SYMBOL(kern_path);
2426 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2427 * @dentry: pointer to dentry of the base directory
2428 * @mnt: pointer to vfs mount of the base directory
2429 * @name: pointer to file name
2430 * @flags: lookup flags
2431 * @path: pointer to struct path to fill
2433 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2434 const char *name, unsigned int flags,
2437 struct path root = {.mnt = mnt, .dentry = dentry};
2438 /* the first argument of filename_lookup() is ignored with root */
2439 return filename_lookup(AT_FDCWD, getname_kernel(name),
2440 flags , path, &root);
2442 EXPORT_SYMBOL(vfs_path_lookup);
2444 static int lookup_one_len_common(const char *name, struct dentry *base,
2445 int len, struct qstr *this)
2449 this->hash = full_name_hash(base, name, len);
2453 if (unlikely(name[0] == '.')) {
2454 if (len < 2 || (len == 2 && name[1] == '.'))
2459 unsigned int c = *(const unsigned char *)name++;
2460 if (c == '/' || c == '\0')
2464 * See if the low-level filesystem might want
2465 * to use its own hash..
2467 if (base->d_flags & DCACHE_OP_HASH) {
2468 int err = base->d_op->d_hash(base, this);
2473 return inode_permission(base->d_inode, MAY_EXEC);
2477 * try_lookup_one_len - filesystem helper to lookup single pathname component
2478 * @name: pathname component to lookup
2479 * @base: base directory to lookup from
2480 * @len: maximum length @len should be interpreted to
2482 * Look up a dentry by name in the dcache, returning NULL if it does not
2483 * currently exist. The function does not try to create a dentry.
2485 * Note that this routine is purely a helper for filesystem usage and should
2486 * not be called by generic code.
2488 * The caller must hold base->i_mutex.
2490 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2495 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2497 err = lookup_one_len_common(name, base, len, &this);
2499 return ERR_PTR(err);
2501 return lookup_dcache(&this, base, 0);
2503 EXPORT_SYMBOL(try_lookup_one_len);
2506 * lookup_one_len - filesystem helper to lookup single pathname component
2507 * @name: pathname component to lookup
2508 * @base: base directory to lookup from
2509 * @len: maximum length @len should be interpreted to
2511 * Note that this routine is purely a helper for filesystem usage and should
2512 * not be called by generic code.
2514 * The caller must hold base->i_mutex.
2516 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2518 struct dentry *dentry;
2522 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2524 err = lookup_one_len_common(name, base, len, &this);
2526 return ERR_PTR(err);
2528 dentry = lookup_dcache(&this, base, 0);
2529 return dentry ? dentry : __lookup_slow(&this, base, 0);
2531 EXPORT_SYMBOL(lookup_one_len);
2534 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2535 * @name: pathname component to lookup
2536 * @base: base directory to lookup from
2537 * @len: maximum length @len should be interpreted to
2539 * Note that this routine is purely a helper for filesystem usage and should
2540 * not be called by generic code.
2542 * Unlike lookup_one_len, it should be called without the parent
2543 * i_mutex held, and will take the i_mutex itself if necessary.
2545 struct dentry *lookup_one_len_unlocked(const char *name,
2546 struct dentry *base, int len)
2552 err = lookup_one_len_common(name, base, len, &this);
2554 return ERR_PTR(err);
2556 ret = lookup_dcache(&this, base, 0);
2558 ret = lookup_slow(&this, base, 0);
2561 EXPORT_SYMBOL(lookup_one_len_unlocked);
2564 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2565 * on negatives. Returns known positive or ERR_PTR(); that's what
2566 * most of the users want. Note that pinned negative with unlocked parent
2567 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2568 * need to be very careful; pinned positives have ->d_inode stable, so
2569 * this one avoids such problems.
2571 struct dentry *lookup_positive_unlocked(const char *name,
2572 struct dentry *base, int len)
2574 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2575 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2577 ret = ERR_PTR(-ENOENT);
2581 EXPORT_SYMBOL(lookup_positive_unlocked);
2583 #ifdef CONFIG_UNIX98_PTYS
2584 int path_pts(struct path *path)
2586 /* Find something mounted on "pts" in the same directory as
2589 struct dentry *child, *parent;
2593 ret = path_parent_directory(path);
2597 parent = path->dentry;
2600 child = d_hash_and_lookup(parent, &this);
2604 path->dentry = child;
2611 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2612 struct path *path, int *empty)
2614 return filename_lookup(dfd, getname_flags(name, flags, empty),
2617 EXPORT_SYMBOL(user_path_at_empty);
2620 * path_mountpoint - look up a path to be umounted
2621 * @nd: lookup context
2622 * @flags: lookup flags
2623 * @path: pointer to container for result
2625 * Look up the given name, but don't attempt to revalidate the last component.
2626 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2629 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2631 const char *s = path_init(nd, flags);
2634 while (!(err = link_path_walk(s, nd)) &&
2635 (err = lookup_last(nd)) > 0) {
2636 s = trailing_symlink(nd);
2638 if (!err && (nd->flags & LOOKUP_RCU))
2639 err = unlazy_walk(nd);
2641 err = handle_lookup_down(nd);
2644 nd->path.mnt = NULL;
2645 nd->path.dentry = NULL;
2652 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2655 struct nameidata nd;
2658 return PTR_ERR(name);
2659 set_nameidata(&nd, dfd, name);
2660 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2661 if (unlikely(error == -ECHILD))
2662 error = path_mountpoint(&nd, flags, path);
2663 if (unlikely(error == -ESTALE))
2664 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2666 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
2667 restore_nameidata();
2673 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2674 * @dfd: directory file descriptor
2675 * @name: pathname from userland
2676 * @flags: lookup flags
2677 * @path: pointer to container to hold result
2679 * A umount is a special case for path walking. We're not actually interested
2680 * in the inode in this situation, and ESTALE errors can be a problem. We
2681 * simply want track down the dentry and vfsmount attached at the mountpoint
2682 * and avoid revalidating the last component.
2684 * Returns 0 and populates "path" on success.
2687 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2690 return filename_mountpoint(dfd, getname(name), path, flags);
2694 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2697 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2699 EXPORT_SYMBOL(kern_path_mountpoint);
2701 int __check_sticky(struct inode *dir, struct inode *inode)
2703 kuid_t fsuid = current_fsuid();
2705 if (uid_eq(inode->i_uid, fsuid))
2707 if (uid_eq(dir->i_uid, fsuid))
2709 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2711 EXPORT_SYMBOL(__check_sticky);
2714 * Check whether we can remove a link victim from directory dir, check
2715 * whether the type of victim is right.
2716 * 1. We can't do it if dir is read-only (done in permission())
2717 * 2. We should have write and exec permissions on dir
2718 * 3. We can't remove anything from append-only dir
2719 * 4. We can't do anything with immutable dir (done in permission())
2720 * 5. If the sticky bit on dir is set we should either
2721 * a. be owner of dir, or
2722 * b. be owner of victim, or
2723 * c. have CAP_FOWNER capability
2724 * 6. If the victim is append-only or immutable we can't do antyhing with
2725 * links pointing to it.
2726 * 7. If the victim has an unknown uid or gid we can't change the inode.
2727 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2728 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2729 * 10. We can't remove a root or mountpoint.
2730 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2731 * nfs_async_unlink().
2733 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2735 struct inode *inode = d_backing_inode(victim);
2738 if (d_is_negative(victim))
2742 BUG_ON(victim->d_parent->d_inode != dir);
2744 /* Inode writeback is not safe when the uid or gid are invalid. */
2745 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2748 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2750 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2756 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2757 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2760 if (!d_is_dir(victim))
2762 if (IS_ROOT(victim))
2764 } else if (d_is_dir(victim))
2766 if (IS_DEADDIR(dir))
2768 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2773 /* Check whether we can create an object with dentry child in directory
2775 * 1. We can't do it if child already exists (open has special treatment for
2776 * this case, but since we are inlined it's OK)
2777 * 2. We can't do it if dir is read-only (done in permission())
2778 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2779 * 4. We should have write and exec permissions on dir
2780 * 5. We can't do it if dir is immutable (done in permission())
2782 static inline int may_create(struct inode *dir, struct dentry *child)
2784 struct user_namespace *s_user_ns;
2785 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2788 if (IS_DEADDIR(dir))
2790 s_user_ns = dir->i_sb->s_user_ns;
2791 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2792 !kgid_has_mapping(s_user_ns, current_fsgid()))
2794 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2798 * p1 and p2 should be directories on the same fs.
2800 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2805 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2809 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2811 p = d_ancestor(p2, p1);
2813 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2814 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2818 p = d_ancestor(p1, p2);
2820 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2821 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2825 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2826 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2829 EXPORT_SYMBOL(lock_rename);
2831 void unlock_rename(struct dentry *p1, struct dentry *p2)
2833 inode_unlock(p1->d_inode);
2835 inode_unlock(p2->d_inode);
2836 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2839 EXPORT_SYMBOL(unlock_rename);
2841 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2844 int error = may_create(dir, dentry);
2848 if (!dir->i_op->create)
2849 return -EACCES; /* shouldn't it be ENOSYS? */
2852 error = security_inode_create(dir, dentry, mode);
2855 error = dir->i_op->create(dir, dentry, mode, want_excl);
2857 fsnotify_create(dir, dentry);
2860 EXPORT_SYMBOL(vfs_create);
2862 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2863 int (*f)(struct dentry *, umode_t, void *),
2866 struct inode *dir = dentry->d_parent->d_inode;
2867 int error = may_create(dir, dentry);
2873 error = security_inode_create(dir, dentry, mode);
2876 error = f(dentry, mode, arg);
2878 fsnotify_create(dir, dentry);
2881 EXPORT_SYMBOL(vfs_mkobj);
2883 bool may_open_dev(const struct path *path)
2885 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2886 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2889 static int may_open(const struct path *path, int acc_mode, int flag)
2891 struct dentry *dentry = path->dentry;
2892 struct inode *inode = dentry->d_inode;
2898 switch (inode->i_mode & S_IFMT) {
2902 if (acc_mode & MAY_WRITE)
2907 if (!may_open_dev(path))
2916 error = inode_permission(inode, MAY_OPEN | acc_mode);
2921 * An append-only file must be opened in append mode for writing.
2923 if (IS_APPEND(inode)) {
2924 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2930 /* O_NOATIME can only be set by the owner or superuser */
2931 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2937 static int handle_truncate(struct file *filp)
2939 const struct path *path = &filp->f_path;
2940 struct inode *inode = path->dentry->d_inode;
2941 int error = get_write_access(inode);
2945 * Refuse to truncate files with mandatory locks held on them.
2947 error = locks_verify_locked(filp);
2949 error = security_path_truncate(path);
2951 error = do_truncate(path->dentry, 0,
2952 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2955 put_write_access(inode);
2959 static inline int open_to_namei_flags(int flag)
2961 if ((flag & O_ACCMODE) == 3)
2966 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2968 struct user_namespace *s_user_ns;
2969 int error = security_path_mknod(dir, dentry, mode, 0);
2973 s_user_ns = dir->dentry->d_sb->s_user_ns;
2974 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2975 !kgid_has_mapping(s_user_ns, current_fsgid()))
2978 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2982 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2986 * Attempt to atomically look up, create and open a file from a negative
2989 * Returns 0 if successful. The file will have been created and attached to
2990 * @file by the filesystem calling finish_open().
2992 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
2993 * be set. The caller will need to perform the open themselves. @path will
2994 * have been updated to point to the new dentry. This may be negative.
2996 * Returns an error code otherwise.
2998 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2999 struct path *path, struct file *file,
3000 const struct open_flags *op,
3001 int open_flag, umode_t mode)
3003 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3004 struct inode *dir = nd->path.dentry->d_inode;
3007 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3008 open_flag &= ~O_TRUNC;
3010 if (nd->flags & LOOKUP_DIRECTORY)
3011 open_flag |= O_DIRECTORY;
3013 file->f_path.dentry = DENTRY_NOT_SET;
3014 file->f_path.mnt = nd->path.mnt;
3015 error = dir->i_op->atomic_open(dir, dentry, file,
3016 open_to_namei_flags(open_flag), mode);
3017 d_lookup_done(dentry);
3019 if (file->f_mode & FMODE_OPENED) {
3021 * We didn't have the inode before the open, so check open
3024 int acc_mode = op->acc_mode;
3025 if (file->f_mode & FMODE_CREATED) {
3026 WARN_ON(!(open_flag & O_CREAT));
3027 fsnotify_create(dir, dentry);
3030 error = may_open(&file->f_path, acc_mode, open_flag);
3031 if (WARN_ON(error > 0))
3033 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3036 if (file->f_path.dentry) {
3038 dentry = file->f_path.dentry;
3040 if (file->f_mode & FMODE_CREATED)
3041 fsnotify_create(dir, dentry);
3042 if (unlikely(d_is_negative(dentry))) {
3045 path->dentry = dentry;
3046 path->mnt = nd->path.mnt;
3056 * Look up and maybe create and open the last component.
3058 * Must be called with parent locked (exclusive in O_CREAT case).
3060 * Returns 0 on success, that is, if
3061 * the file was successfully atomically created (if necessary) and opened, or
3062 * the file was not completely opened at this time, though lookups and
3063 * creations were performed.
3064 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3065 * In the latter case dentry returned in @path might be negative if O_CREAT
3066 * hadn't been specified.
3068 * An error code is returned on failure.
3070 static int lookup_open(struct nameidata *nd, struct path *path,
3072 const struct open_flags *op,
3075 struct dentry *dir = nd->path.dentry;
3076 struct inode *dir_inode = dir->d_inode;
3077 int open_flag = op->open_flag;
3078 struct dentry *dentry;
3079 int error, create_error = 0;
3080 umode_t mode = op->mode;
3081 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3083 if (unlikely(IS_DEADDIR(dir_inode)))
3086 file->f_mode &= ~FMODE_CREATED;
3087 dentry = d_lookup(dir, &nd->last);
3090 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3092 return PTR_ERR(dentry);
3094 if (d_in_lookup(dentry))
3097 error = d_revalidate(dentry, nd->flags);
3098 if (likely(error > 0))
3102 d_invalidate(dentry);
3106 if (dentry->d_inode) {
3107 /* Cached positive dentry: will open in f_op->open */
3112 * Checking write permission is tricky, bacuse we don't know if we are
3113 * going to actually need it: O_CREAT opens should work as long as the
3114 * file exists. But checking existence breaks atomicity. The trick is
3115 * to check access and if not granted clear O_CREAT from the flags.
3117 * Another problem is returing the "right" error value (e.g. for an
3118 * O_EXCL open we want to return EEXIST not EROFS).
3120 if (open_flag & O_CREAT) {
3121 if (!IS_POSIXACL(dir->d_inode))
3122 mode &= ~current_umask();
3123 if (unlikely(!got_write)) {
3124 create_error = -EROFS;
3125 open_flag &= ~O_CREAT;
3126 if (open_flag & (O_EXCL | O_TRUNC))
3128 /* No side effects, safe to clear O_CREAT */
3130 create_error = may_o_create(&nd->path, dentry, mode);
3132 open_flag &= ~O_CREAT;
3133 if (open_flag & O_EXCL)
3137 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3138 unlikely(!got_write)) {
3140 * No O_CREATE -> atomicity not a requirement -> fall
3141 * back to lookup + open
3146 if (dir_inode->i_op->atomic_open) {
3147 error = atomic_open(nd, dentry, path, file, op, open_flag,
3149 if (unlikely(error == -ENOENT) && create_error)
3150 error = create_error;
3155 if (d_in_lookup(dentry)) {
3156 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3158 d_lookup_done(dentry);
3159 if (unlikely(res)) {
3161 error = PTR_ERR(res);
3169 /* Negative dentry, just create the file */
3170 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3171 file->f_mode |= FMODE_CREATED;
3172 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3173 if (!dir_inode->i_op->create) {
3177 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3178 open_flag & O_EXCL);
3181 fsnotify_create(dir_inode, dentry);
3183 if (unlikely(create_error) && !dentry->d_inode) {
3184 error = create_error;
3188 path->dentry = dentry;
3189 path->mnt = nd->path.mnt;
3198 * Handle the last step of open()
3200 static int do_last(struct nameidata *nd,
3201 struct file *file, const struct open_flags *op)
3203 struct dentry *dir = nd->path.dentry;
3204 int open_flag = op->open_flag;
3205 bool will_truncate = (open_flag & O_TRUNC) != 0;
3206 bool got_write = false;
3207 int acc_mode = op->acc_mode;
3209 struct inode *inode;
3213 nd->flags &= ~LOOKUP_PARENT;
3214 nd->flags |= op->intent;
3216 if (nd->last_type != LAST_NORM) {
3217 error = handle_dots(nd, nd->last_type);
3218 if (unlikely(error))
3223 if (!(open_flag & O_CREAT)) {
3224 if (nd->last.name[nd->last.len])
3225 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3226 /* we _can_ be in RCU mode here */
3227 error = lookup_fast(nd, &path, &inode, &seq);
3228 if (likely(error > 0))
3234 BUG_ON(nd->inode != dir->d_inode);
3235 BUG_ON(nd->flags & LOOKUP_RCU);
3237 /* create side of things */
3239 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3240 * has been cleared when we got to the last component we are
3243 error = complete_walk(nd);
3247 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3248 /* trailing slashes? */
3249 if (unlikely(nd->last.name[nd->last.len]))
3253 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3254 error = mnt_want_write(nd->path.mnt);
3258 * do _not_ fail yet - we might not need that or fail with
3259 * a different error; let lookup_open() decide; we'll be
3260 * dropping this one anyway.
3263 if (open_flag & O_CREAT)
3264 inode_lock(dir->d_inode);
3266 inode_lock_shared(dir->d_inode);
3267 error = lookup_open(nd, &path, file, op, got_write);
3268 if (open_flag & O_CREAT)
3269 inode_unlock(dir->d_inode);
3271 inode_unlock_shared(dir->d_inode);
3276 if (file->f_mode & FMODE_OPENED) {
3277 if ((file->f_mode & FMODE_CREATED) ||
3278 !S_ISREG(file_inode(file)->i_mode))
3279 will_truncate = false;
3281 audit_inode(nd->name, file->f_path.dentry, 0);
3285 if (file->f_mode & FMODE_CREATED) {
3286 /* Don't check for write permission, don't truncate */
3287 open_flag &= ~O_TRUNC;
3288 will_truncate = false;
3290 path_to_nameidata(&path, nd);
3291 goto finish_open_created;
3295 * If atomic_open() acquired write access it is dropped now due to
3296 * possible mount and symlink following (this might be optimized away if
3300 mnt_drop_write(nd->path.mnt);
3304 error = follow_managed(&path, nd);
3305 if (unlikely(error < 0))
3309 * create/update audit record if it already exists.
3311 audit_inode(nd->name, path.dentry, 0);
3313 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3314 path_to_nameidata(&path, nd);
3318 seq = 0; /* out of RCU mode, so the value doesn't matter */
3319 inode = d_backing_inode(path.dentry);
3321 error = step_into(nd, &path, 0, inode, seq);
3322 if (unlikely(error))
3325 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3326 error = complete_walk(nd);
3329 audit_inode(nd->name, nd->path.dentry, 0);
3330 if (open_flag & O_CREAT) {
3332 if (d_is_dir(nd->path.dentry))
3334 error = may_create_in_sticky(dir,
3335 d_backing_inode(nd->path.dentry));
3336 if (unlikely(error))
3340 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3342 if (!d_is_reg(nd->path.dentry))
3343 will_truncate = false;
3345 if (will_truncate) {
3346 error = mnt_want_write(nd->path.mnt);
3351 finish_open_created:
3352 error = may_open(&nd->path, acc_mode, open_flag);
3355 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3356 error = vfs_open(&nd->path, file);
3360 error = ima_file_check(file, op->acc_mode);
3361 if (!error && will_truncate)
3362 error = handle_truncate(file);
3364 if (unlikely(error > 0)) {
3369 mnt_drop_write(nd->path.mnt);
3373 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3375 struct dentry *child = NULL;
3376 struct inode *dir = dentry->d_inode;
3377 struct inode *inode;
3380 /* we want directory to be writable */
3381 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3384 error = -EOPNOTSUPP;
3385 if (!dir->i_op->tmpfile)
3388 child = d_alloc(dentry, &slash_name);
3389 if (unlikely(!child))
3391 error = dir->i_op->tmpfile(dir, child, mode);
3395 inode = child->d_inode;
3396 if (unlikely(!inode))
3398 if (!(open_flag & O_EXCL)) {
3399 spin_lock(&inode->i_lock);
3400 inode->i_state |= I_LINKABLE;
3401 spin_unlock(&inode->i_lock);
3403 ima_post_create_tmpfile(inode);
3408 return ERR_PTR(error);
3410 EXPORT_SYMBOL(vfs_tmpfile);
3412 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3413 const struct open_flags *op,
3416 struct dentry *child;
3418 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3419 if (unlikely(error))
3421 error = mnt_want_write(path.mnt);
3422 if (unlikely(error))
3424 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3425 error = PTR_ERR(child);
3429 path.dentry = child;
3430 audit_inode(nd->name, child, 0);
3431 /* Don't check for other permissions, the inode was just created */
3432 error = may_open(&path, 0, op->open_flag);
3435 file->f_path.mnt = path.mnt;
3436 error = finish_open(file, child, NULL);
3438 mnt_drop_write(path.mnt);
3444 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3447 int error = path_lookupat(nd, flags, &path);
3449 audit_inode(nd->name, path.dentry, 0);
3450 error = vfs_open(&path, file);
3456 static struct file *path_openat(struct nameidata *nd,
3457 const struct open_flags *op, unsigned flags)
3462 file = alloc_empty_file(op->open_flag, current_cred());
3466 if (unlikely(file->f_flags & __O_TMPFILE)) {
3467 error = do_tmpfile(nd, flags, op, file);
3468 } else if (unlikely(file->f_flags & O_PATH)) {
3469 error = do_o_path(nd, flags, file);
3471 const char *s = path_init(nd, flags);
3472 while (!(error = link_path_walk(s, nd)) &&
3473 (error = do_last(nd, file, op)) > 0) {
3474 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3475 s = trailing_symlink(nd);
3479 if (likely(!error)) {
3480 if (likely(file->f_mode & FMODE_OPENED))
3486 if (error == -EOPENSTALE) {
3487 if (flags & LOOKUP_RCU)
3492 return ERR_PTR(error);
3495 struct file *do_filp_open(int dfd, struct filename *pathname,
3496 const struct open_flags *op)
3498 struct nameidata nd;
3499 int flags = op->lookup_flags;
3502 set_nameidata(&nd, dfd, pathname);
3503 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3504 if (unlikely(filp == ERR_PTR(-ECHILD)))
3505 filp = path_openat(&nd, op, flags);
3506 if (unlikely(filp == ERR_PTR(-ESTALE)))
3507 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3508 restore_nameidata();
3512 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3513 const char *name, const struct open_flags *op)
3515 struct nameidata nd;
3517 struct filename *filename;
3518 int flags = op->lookup_flags | LOOKUP_ROOT;
3521 nd.root.dentry = dentry;
3523 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3524 return ERR_PTR(-ELOOP);
3526 filename = getname_kernel(name);
3527 if (IS_ERR(filename))
3528 return ERR_CAST(filename);
3530 set_nameidata(&nd, -1, filename);
3531 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3532 if (unlikely(file == ERR_PTR(-ECHILD)))
3533 file = path_openat(&nd, op, flags);
3534 if (unlikely(file == ERR_PTR(-ESTALE)))
3535 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3536 restore_nameidata();
3541 static struct dentry *filename_create(int dfd, struct filename *name,
3542 struct path *path, unsigned int lookup_flags)
3544 struct dentry *dentry = ERR_PTR(-EEXIST);
3549 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3552 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3553 * other flags passed in are ignored!
3555 lookup_flags &= LOOKUP_REVAL;
3557 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3559 return ERR_CAST(name);
3562 * Yucky last component or no last component at all?
3563 * (foo/., foo/.., /////)
3565 if (unlikely(type != LAST_NORM))
3568 /* don't fail immediately if it's r/o, at least try to report other errors */
3569 err2 = mnt_want_write(path->mnt);
3571 * Do the final lookup.
3573 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3574 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3575 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3580 if (d_is_positive(dentry))
3584 * Special case - lookup gave negative, but... we had foo/bar/
3585 * From the vfs_mknod() POV we just have a negative dentry -
3586 * all is fine. Let's be bastards - you had / on the end, you've
3587 * been asking for (non-existent) directory. -ENOENT for you.
3589 if (unlikely(!is_dir && last.name[last.len])) {
3593 if (unlikely(err2)) {
3601 dentry = ERR_PTR(error);
3603 inode_unlock(path->dentry->d_inode);
3605 mnt_drop_write(path->mnt);
3612 struct dentry *kern_path_create(int dfd, const char *pathname,
3613 struct path *path, unsigned int lookup_flags)
3615 return filename_create(dfd, getname_kernel(pathname),
3616 path, lookup_flags);
3618 EXPORT_SYMBOL(kern_path_create);
3620 void done_path_create(struct path *path, struct dentry *dentry)
3623 inode_unlock(path->dentry->d_inode);
3624 mnt_drop_write(path->mnt);
3627 EXPORT_SYMBOL(done_path_create);
3629 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3630 struct path *path, unsigned int lookup_flags)
3632 return filename_create(dfd, getname(pathname), path, lookup_flags);
3634 EXPORT_SYMBOL(user_path_create);
3636 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3638 int error = may_create(dir, dentry);
3643 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3646 if (!dir->i_op->mknod)
3649 error = devcgroup_inode_mknod(mode, dev);
3653 error = security_inode_mknod(dir, dentry, mode, dev);
3657 error = dir->i_op->mknod(dir, dentry, mode, dev);
3659 fsnotify_create(dir, dentry);
3662 EXPORT_SYMBOL(vfs_mknod);
3664 static int may_mknod(umode_t mode)
3666 switch (mode & S_IFMT) {
3672 case 0: /* zero mode translates to S_IFREG */
3681 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3684 struct dentry *dentry;
3687 unsigned int lookup_flags = 0;
3689 error = may_mknod(mode);
3693 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3695 return PTR_ERR(dentry);
3697 if (!IS_POSIXACL(path.dentry->d_inode))
3698 mode &= ~current_umask();
3699 error = security_path_mknod(&path, dentry, mode, dev);
3702 switch (mode & S_IFMT) {
3703 case 0: case S_IFREG:
3704 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3706 ima_post_path_mknod(dentry);
3708 case S_IFCHR: case S_IFBLK:
3709 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3710 new_decode_dev(dev));
3712 case S_IFIFO: case S_IFSOCK:
3713 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3717 done_path_create(&path, dentry);
3718 if (retry_estale(error, lookup_flags)) {
3719 lookup_flags |= LOOKUP_REVAL;
3725 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3728 return do_mknodat(dfd, filename, mode, dev);
3731 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3733 return do_mknodat(AT_FDCWD, filename, mode, dev);
3736 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3738 int error = may_create(dir, dentry);
3739 unsigned max_links = dir->i_sb->s_max_links;
3744 if (!dir->i_op->mkdir)
3747 mode &= (S_IRWXUGO|S_ISVTX);
3748 error = security_inode_mkdir(dir, dentry, mode);
3752 if (max_links && dir->i_nlink >= max_links)
3755 error = dir->i_op->mkdir(dir, dentry, mode);
3757 fsnotify_mkdir(dir, dentry);
3760 EXPORT_SYMBOL(vfs_mkdir);
3762 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3764 struct dentry *dentry;
3767 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3770 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3772 return PTR_ERR(dentry);
3774 if (!IS_POSIXACL(path.dentry->d_inode))
3775 mode &= ~current_umask();
3776 error = security_path_mkdir(&path, dentry, mode);
3778 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3779 done_path_create(&path, dentry);
3780 if (retry_estale(error, lookup_flags)) {
3781 lookup_flags |= LOOKUP_REVAL;
3787 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3789 return do_mkdirat(dfd, pathname, mode);
3792 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3794 return do_mkdirat(AT_FDCWD, pathname, mode);
3797 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3799 int error = may_delete(dir, dentry, 1);
3804 if (!dir->i_op->rmdir)
3808 inode_lock(dentry->d_inode);
3811 if (is_local_mountpoint(dentry))
3814 error = security_inode_rmdir(dir, dentry);
3818 error = dir->i_op->rmdir(dir, dentry);
3822 shrink_dcache_parent(dentry);
3823 dentry->d_inode->i_flags |= S_DEAD;
3825 detach_mounts(dentry);
3826 fsnotify_rmdir(dir, dentry);
3829 inode_unlock(dentry->d_inode);
3835 EXPORT_SYMBOL(vfs_rmdir);
3837 long do_rmdir(int dfd, const char __user *pathname)
3840 struct filename *name;
3841 struct dentry *dentry;
3845 unsigned int lookup_flags = 0;
3847 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3848 &path, &last, &type);
3850 return PTR_ERR(name);
3864 error = mnt_want_write(path.mnt);
3868 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3869 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3870 error = PTR_ERR(dentry);
3873 if (!dentry->d_inode) {
3877 error = security_path_rmdir(&path, dentry);
3880 error = vfs_rmdir(path.dentry->d_inode, dentry);
3884 inode_unlock(path.dentry->d_inode);
3885 mnt_drop_write(path.mnt);
3889 if (retry_estale(error, lookup_flags)) {
3890 lookup_flags |= LOOKUP_REVAL;
3896 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3898 return do_rmdir(AT_FDCWD, pathname);
3902 * vfs_unlink - unlink a filesystem object
3903 * @dir: parent directory
3905 * @delegated_inode: returns victim inode, if the inode is delegated.
3907 * The caller must hold dir->i_mutex.
3909 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3910 * return a reference to the inode in delegated_inode. The caller
3911 * should then break the delegation on that inode and retry. Because
3912 * breaking a delegation may take a long time, the caller should drop
3913 * dir->i_mutex before doing so.
3915 * Alternatively, a caller may pass NULL for delegated_inode. This may
3916 * be appropriate for callers that expect the underlying filesystem not
3917 * to be NFS exported.
3919 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3921 struct inode *target = dentry->d_inode;
3922 int error = may_delete(dir, dentry, 0);
3927 if (!dir->i_op->unlink)
3931 if (is_local_mountpoint(dentry))
3934 error = security_inode_unlink(dir, dentry);
3936 error = try_break_deleg(target, delegated_inode);
3939 error = dir->i_op->unlink(dir, dentry);
3942 detach_mounts(dentry);
3943 fsnotify_unlink(dir, dentry);
3948 inode_unlock(target);
3950 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3951 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3952 fsnotify_link_count(target);
3958 EXPORT_SYMBOL(vfs_unlink);
3961 * Make sure that the actual truncation of the file will occur outside its
3962 * directory's i_mutex. Truncate can take a long time if there is a lot of
3963 * writeout happening, and we don't want to prevent access to the directory
3964 * while waiting on the I/O.
3966 long do_unlinkat(int dfd, struct filename *name)
3969 struct dentry *dentry;
3973 struct inode *inode = NULL;
3974 struct inode *delegated_inode = NULL;
3975 unsigned int lookup_flags = 0;
3977 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3979 return PTR_ERR(name);
3982 if (type != LAST_NORM)
3985 error = mnt_want_write(path.mnt);
3989 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3990 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3991 error = PTR_ERR(dentry);
3992 if (!IS_ERR(dentry)) {
3993 /* Why not before? Because we want correct error value */
3994 if (last.name[last.len])
3996 inode = dentry->d_inode;
3997 if (d_is_negative(dentry))
4000 error = security_path_unlink(&path, dentry);
4003 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4007 inode_unlock(path.dentry->d_inode);
4009 iput(inode); /* truncate the inode here */
4011 if (delegated_inode) {
4012 error = break_deleg_wait(&delegated_inode);
4016 mnt_drop_write(path.mnt);
4019 if (retry_estale(error, lookup_flags)) {
4020 lookup_flags |= LOOKUP_REVAL;
4028 if (d_is_negative(dentry))
4030 else if (d_is_dir(dentry))
4037 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4039 if ((flag & ~AT_REMOVEDIR) != 0)
4042 if (flag & AT_REMOVEDIR)
4043 return do_rmdir(dfd, pathname);
4045 return do_unlinkat(dfd, getname(pathname));
4048 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4050 return do_unlinkat(AT_FDCWD, getname(pathname));
4053 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4055 int error = may_create(dir, dentry);
4060 if (!dir->i_op->symlink)
4063 error = security_inode_symlink(dir, dentry, oldname);
4067 error = dir->i_op->symlink(dir, dentry, oldname);
4069 fsnotify_create(dir, dentry);
4072 EXPORT_SYMBOL(vfs_symlink);
4074 long do_symlinkat(const char __user *oldname, int newdfd,
4075 const char __user *newname)
4078 struct filename *from;
4079 struct dentry *dentry;
4081 unsigned int lookup_flags = 0;
4083 from = getname(oldname);
4085 return PTR_ERR(from);
4087 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4088 error = PTR_ERR(dentry);
4092 error = security_path_symlink(&path, dentry, from->name);
4094 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4095 done_path_create(&path, dentry);
4096 if (retry_estale(error, lookup_flags)) {
4097 lookup_flags |= LOOKUP_REVAL;
4105 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4106 int, newdfd, const char __user *, newname)
4108 return do_symlinkat(oldname, newdfd, newname);
4111 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4113 return do_symlinkat(oldname, AT_FDCWD, newname);
4117 * vfs_link - create a new link
4118 * @old_dentry: object to be linked
4120 * @new_dentry: where to create the new link
4121 * @delegated_inode: returns inode needing a delegation break
4123 * The caller must hold dir->i_mutex
4125 * If vfs_link discovers a delegation on the to-be-linked file in need
4126 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4127 * inode in delegated_inode. The caller should then break the delegation
4128 * and retry. Because breaking a delegation may take a long time, the
4129 * caller should drop the i_mutex before doing so.
4131 * Alternatively, a caller may pass NULL for delegated_inode. This may
4132 * be appropriate for callers that expect the underlying filesystem not
4133 * to be NFS exported.
4135 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4137 struct inode *inode = old_dentry->d_inode;
4138 unsigned max_links = dir->i_sb->s_max_links;
4144 error = may_create(dir, new_dentry);
4148 if (dir->i_sb != inode->i_sb)
4152 * A link to an append-only or immutable file cannot be created.
4154 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4157 * Updating the link count will likely cause i_uid and i_gid to
4158 * be writen back improperly if their true value is unknown to
4161 if (HAS_UNMAPPED_ID(inode))
4163 if (!dir->i_op->link)
4165 if (S_ISDIR(inode->i_mode))
4168 error = security_inode_link(old_dentry, dir, new_dentry);
4173 /* Make sure we don't allow creating hardlink to an unlinked file */
4174 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4176 else if (max_links && inode->i_nlink >= max_links)
4179 error = try_break_deleg(inode, delegated_inode);
4181 error = dir->i_op->link(old_dentry, dir, new_dentry);
4184 if (!error && (inode->i_state & I_LINKABLE)) {
4185 spin_lock(&inode->i_lock);
4186 inode->i_state &= ~I_LINKABLE;
4187 spin_unlock(&inode->i_lock);
4189 inode_unlock(inode);
4191 fsnotify_link(dir, inode, new_dentry);
4194 EXPORT_SYMBOL(vfs_link);
4197 * Hardlinks are often used in delicate situations. We avoid
4198 * security-related surprises by not following symlinks on the
4201 * We don't follow them on the oldname either to be compatible
4202 * with linux 2.0, and to avoid hard-linking to directories
4203 * and other special files. --ADM
4205 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4206 const char __user *newname, int flags)
4208 struct dentry *new_dentry;
4209 struct path old_path, new_path;
4210 struct inode *delegated_inode = NULL;
4214 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4217 * To use null names we require CAP_DAC_READ_SEARCH
4218 * This ensures that not everyone will be able to create
4219 * handlink using the passed filedescriptor.
4221 if (flags & AT_EMPTY_PATH) {
4222 if (!capable(CAP_DAC_READ_SEARCH))
4227 if (flags & AT_SYMLINK_FOLLOW)
4228 how |= LOOKUP_FOLLOW;
4230 error = user_path_at(olddfd, oldname, how, &old_path);
4234 new_dentry = user_path_create(newdfd, newname, &new_path,
4235 (how & LOOKUP_REVAL));
4236 error = PTR_ERR(new_dentry);
4237 if (IS_ERR(new_dentry))
4241 if (old_path.mnt != new_path.mnt)
4243 error = may_linkat(&old_path);
4244 if (unlikely(error))
4246 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4249 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4251 done_path_create(&new_path, new_dentry);
4252 if (delegated_inode) {
4253 error = break_deleg_wait(&delegated_inode);
4255 path_put(&old_path);
4259 if (retry_estale(error, how)) {
4260 path_put(&old_path);
4261 how |= LOOKUP_REVAL;
4265 path_put(&old_path);
4270 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4271 int, newdfd, const char __user *, newname, int, flags)
4273 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4276 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4278 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4282 * vfs_rename - rename a filesystem object
4283 * @old_dir: parent of source
4284 * @old_dentry: source
4285 * @new_dir: parent of destination
4286 * @new_dentry: destination
4287 * @delegated_inode: returns an inode needing a delegation break
4288 * @flags: rename flags
4290 * The caller must hold multiple mutexes--see lock_rename()).
4292 * If vfs_rename discovers a delegation in need of breaking at either
4293 * the source or destination, it will return -EWOULDBLOCK and return a
4294 * reference to the inode in delegated_inode. The caller should then
4295 * break the delegation and retry. Because breaking a delegation may
4296 * take a long time, the caller should drop all locks before doing
4299 * Alternatively, a caller may pass NULL for delegated_inode. This may
4300 * be appropriate for callers that expect the underlying filesystem not
4301 * to be NFS exported.
4303 * The worst of all namespace operations - renaming directory. "Perverted"
4304 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4307 * a) we can get into loop creation.
4308 * b) race potential - two innocent renames can create a loop together.
4309 * That's where 4.4 screws up. Current fix: serialization on
4310 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4312 * c) we have to lock _four_ objects - parents and victim (if it exists),
4313 * and source (if it is not a directory).
4314 * And that - after we got ->i_mutex on parents (until then we don't know
4315 * whether the target exists). Solution: try to be smart with locking
4316 * order for inodes. We rely on the fact that tree topology may change
4317 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4318 * move will be locked. Thus we can rank directories by the tree
4319 * (ancestors first) and rank all non-directories after them.
4320 * That works since everybody except rename does "lock parent, lookup,
4321 * lock child" and rename is under ->s_vfs_rename_mutex.
4322 * HOWEVER, it relies on the assumption that any object with ->lookup()
4323 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4324 * we'd better make sure that there's no link(2) for them.
4325 * d) conversion from fhandle to dentry may come in the wrong moment - when
4326 * we are removing the target. Solution: we will have to grab ->i_mutex
4327 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4328 * ->i_mutex on parents, which works but leads to some truly excessive
4331 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4332 struct inode *new_dir, struct dentry *new_dentry,
4333 struct inode **delegated_inode, unsigned int flags)
4336 bool is_dir = d_is_dir(old_dentry);
4337 struct inode *source = old_dentry->d_inode;
4338 struct inode *target = new_dentry->d_inode;
4339 bool new_is_dir = false;
4340 unsigned max_links = new_dir->i_sb->s_max_links;
4341 struct name_snapshot old_name;
4343 if (source == target)
4346 error = may_delete(old_dir, old_dentry, is_dir);
4351 error = may_create(new_dir, new_dentry);
4353 new_is_dir = d_is_dir(new_dentry);
4355 if (!(flags & RENAME_EXCHANGE))
4356 error = may_delete(new_dir, new_dentry, is_dir);
4358 error = may_delete(new_dir, new_dentry, new_is_dir);
4363 if (!old_dir->i_op->rename)
4367 * If we are going to change the parent - check write permissions,
4368 * we'll need to flip '..'.
4370 if (new_dir != old_dir) {
4372 error = inode_permission(source, MAY_WRITE);
4376 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4377 error = inode_permission(target, MAY_WRITE);
4383 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4388 take_dentry_name_snapshot(&old_name, old_dentry);
4390 if (!is_dir || (flags & RENAME_EXCHANGE))
4391 lock_two_nondirectories(source, target);
4396 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4399 if (max_links && new_dir != old_dir) {
4401 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4403 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4404 old_dir->i_nlink >= max_links)
4408 error = try_break_deleg(source, delegated_inode);
4412 if (target && !new_is_dir) {
4413 error = try_break_deleg(target, delegated_inode);
4417 error = old_dir->i_op->rename(old_dir, old_dentry,
4418 new_dir, new_dentry, flags);
4422 if (!(flags & RENAME_EXCHANGE) && target) {
4424 shrink_dcache_parent(new_dentry);
4425 target->i_flags |= S_DEAD;
4427 dont_mount(new_dentry);
4428 detach_mounts(new_dentry);
4430 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4431 if (!(flags & RENAME_EXCHANGE))
4432 d_move(old_dentry, new_dentry);
4434 d_exchange(old_dentry, new_dentry);
4437 if (!is_dir || (flags & RENAME_EXCHANGE))
4438 unlock_two_nondirectories(source, target);
4440 inode_unlock(target);
4443 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4444 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4445 if (flags & RENAME_EXCHANGE) {
4446 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4447 new_is_dir, NULL, new_dentry);
4450 release_dentry_name_snapshot(&old_name);
4454 EXPORT_SYMBOL(vfs_rename);
4456 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4457 const char __user *newname, unsigned int flags)
4459 struct dentry *old_dentry, *new_dentry;
4460 struct dentry *trap;
4461 struct path old_path, new_path;
4462 struct qstr old_last, new_last;
4463 int old_type, new_type;
4464 struct inode *delegated_inode = NULL;
4465 struct filename *from;
4466 struct filename *to;
4467 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4468 bool should_retry = false;
4471 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4474 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4475 (flags & RENAME_EXCHANGE))
4478 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4481 if (flags & RENAME_EXCHANGE)
4485 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4486 &old_path, &old_last, &old_type);
4488 error = PTR_ERR(from);
4492 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4493 &new_path, &new_last, &new_type);
4495 error = PTR_ERR(to);
4500 if (old_path.mnt != new_path.mnt)
4504 if (old_type != LAST_NORM)
4507 if (flags & RENAME_NOREPLACE)
4509 if (new_type != LAST_NORM)
4512 error = mnt_want_write(old_path.mnt);
4517 trap = lock_rename(new_path.dentry, old_path.dentry);
4519 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4520 error = PTR_ERR(old_dentry);
4521 if (IS_ERR(old_dentry))
4523 /* source must exist */
4525 if (d_is_negative(old_dentry))
4527 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4528 error = PTR_ERR(new_dentry);
4529 if (IS_ERR(new_dentry))
4532 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4534 if (flags & RENAME_EXCHANGE) {
4536 if (d_is_negative(new_dentry))
4539 if (!d_is_dir(new_dentry)) {
4541 if (new_last.name[new_last.len])
4545 /* unless the source is a directory trailing slashes give -ENOTDIR */
4546 if (!d_is_dir(old_dentry)) {
4548 if (old_last.name[old_last.len])
4550 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4553 /* source should not be ancestor of target */
4555 if (old_dentry == trap)
4557 /* target should not be an ancestor of source */
4558 if (!(flags & RENAME_EXCHANGE))
4560 if (new_dentry == trap)
4563 error = security_path_rename(&old_path, old_dentry,
4564 &new_path, new_dentry, flags);
4567 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4568 new_path.dentry->d_inode, new_dentry,
4569 &delegated_inode, flags);
4575 unlock_rename(new_path.dentry, old_path.dentry);
4576 if (delegated_inode) {
4577 error = break_deleg_wait(&delegated_inode);
4581 mnt_drop_write(old_path.mnt);
4583 if (retry_estale(error, lookup_flags))
4584 should_retry = true;
4585 path_put(&new_path);
4588 path_put(&old_path);
4591 should_retry = false;
4592 lookup_flags |= LOOKUP_REVAL;
4599 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4600 int, newdfd, const char __user *, newname, unsigned int, flags)
4602 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4605 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4606 int, newdfd, const char __user *, newname)
4608 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4611 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4613 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4616 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4618 int error = may_create(dir, dentry);
4622 if (!dir->i_op->mknod)
4625 return dir->i_op->mknod(dir, dentry,
4626 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4628 EXPORT_SYMBOL(vfs_whiteout);
4630 int readlink_copy(char __user *buffer, int buflen, const char *link)
4632 int len = PTR_ERR(link);
4637 if (len > (unsigned) buflen)
4639 if (copy_to_user(buffer, link, len))
4646 * vfs_readlink - copy symlink body into userspace buffer
4647 * @dentry: dentry on which to get symbolic link
4648 * @buffer: user memory pointer
4649 * @buflen: size of buffer
4651 * Does not touch atime. That's up to the caller if necessary
4653 * Does not call security hook.
4655 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4657 struct inode *inode = d_inode(dentry);
4658 DEFINE_DELAYED_CALL(done);
4662 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4663 if (unlikely(inode->i_op->readlink))
4664 return inode->i_op->readlink(dentry, buffer, buflen);
4666 if (!d_is_symlink(dentry))
4669 spin_lock(&inode->i_lock);
4670 inode->i_opflags |= IOP_DEFAULT_READLINK;
4671 spin_unlock(&inode->i_lock);
4674 link = READ_ONCE(inode->i_link);
4676 link = inode->i_op->get_link(dentry, inode, &done);
4678 return PTR_ERR(link);
4680 res = readlink_copy(buffer, buflen, link);
4681 do_delayed_call(&done);
4684 EXPORT_SYMBOL(vfs_readlink);
4687 * vfs_get_link - get symlink body
4688 * @dentry: dentry on which to get symbolic link
4689 * @done: caller needs to free returned data with this
4691 * Calls security hook and i_op->get_link() on the supplied inode.
4693 * It does not touch atime. That's up to the caller if necessary.
4695 * Does not work on "special" symlinks like /proc/$$/fd/N
4697 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4699 const char *res = ERR_PTR(-EINVAL);
4700 struct inode *inode = d_inode(dentry);
4702 if (d_is_symlink(dentry)) {
4703 res = ERR_PTR(security_inode_readlink(dentry));
4705 res = inode->i_op->get_link(dentry, inode, done);
4709 EXPORT_SYMBOL(vfs_get_link);
4711 /* get the link contents into pagecache */
4712 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4713 struct delayed_call *callback)
4717 struct address_space *mapping = inode->i_mapping;
4720 page = find_get_page(mapping, 0);
4722 return ERR_PTR(-ECHILD);
4723 if (!PageUptodate(page)) {
4725 return ERR_PTR(-ECHILD);
4728 page = read_mapping_page(mapping, 0, NULL);
4732 set_delayed_call(callback, page_put_link, page);
4733 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4734 kaddr = page_address(page);
4735 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4739 EXPORT_SYMBOL(page_get_link);
4741 void page_put_link(void *arg)
4745 EXPORT_SYMBOL(page_put_link);
4747 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4749 DEFINE_DELAYED_CALL(done);
4750 int res = readlink_copy(buffer, buflen,
4751 page_get_link(dentry, d_inode(dentry),
4753 do_delayed_call(&done);
4756 EXPORT_SYMBOL(page_readlink);
4759 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4761 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4763 struct address_space *mapping = inode->i_mapping;
4767 unsigned int flags = 0;
4769 flags |= AOP_FLAG_NOFS;
4772 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4773 flags, &page, &fsdata);
4777 memcpy(page_address(page), symname, len-1);
4779 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4786 mark_inode_dirty(inode);
4791 EXPORT_SYMBOL(__page_symlink);
4793 int page_symlink(struct inode *inode, const char *symname, int len)
4795 return __page_symlink(inode, symname, len,
4796 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4798 EXPORT_SYMBOL(page_symlink);
4800 const struct inode_operations page_symlink_inode_operations = {
4801 .get_link = page_get_link,
4803 EXPORT_SYMBOL(page_symlink_inode_operations);