1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 struct filename *tmp;
227 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
228 if (unlikely(!tmp)) {
230 return ERR_PTR(-ENOMEM);
232 tmp->name = (char *)result;
236 return ERR_PTR(-ENAMETOOLONG);
238 memcpy((char *)result->name, filename, len);
240 result->aname = NULL;
242 audit_getname(result);
247 void putname(struct filename *name)
249 BUG_ON(name->refcnt <= 0);
251 if (--name->refcnt > 0)
254 if (name->name != name->iname) {
255 __putname(name->name);
261 static int check_acl(struct inode *inode, int mask)
263 #ifdef CONFIG_FS_POSIX_ACL
264 struct posix_acl *acl;
266 if (mask & MAY_NOT_BLOCK) {
267 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
270 /* no ->get_acl() calls in RCU mode... */
271 if (is_uncached_acl(acl))
273 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
276 acl = get_acl(inode, ACL_TYPE_ACCESS);
280 int error = posix_acl_permission(inode, acl, mask);
281 posix_acl_release(acl);
290 * This does the basic permission checking
292 static int acl_permission_check(struct inode *inode, int mask)
294 unsigned int mode = inode->i_mode;
296 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
299 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
300 int error = check_acl(inode, mask);
301 if (error != -EAGAIN)
305 if (in_group_p(inode->i_gid))
310 * If the DACs are ok we don't need any capability check.
312 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
318 * generic_permission - check for access rights on a Posix-like filesystem
319 * @inode: inode to check access rights for
320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 * Used to check for read/write/execute permissions on a file.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
327 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
328 * request cannot be satisfied (eg. requires blocking or too much complexity).
329 * It would then be called again in ref-walk mode.
331 int generic_permission(struct inode *inode, int mask)
336 * Do the basic permission checks.
338 ret = acl_permission_check(inode, mask);
342 if (S_ISDIR(inode->i_mode)) {
343 /* DACs are overridable for directories */
344 if (!(mask & MAY_WRITE))
345 if (capable_wrt_inode_uidgid(inode,
346 CAP_DAC_READ_SEARCH))
348 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
354 * Searching includes executable on directories, else just read.
356 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
357 if (mask == MAY_READ)
358 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
361 * Read/write DACs are always overridable.
362 * Executable DACs are overridable when there is
363 * at least one exec bit set.
365 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
366 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
371 EXPORT_SYMBOL(generic_permission);
374 * We _really_ want to just do "generic_permission()" without
375 * even looking at the inode->i_op values. So we keep a cache
376 * flag in inode->i_opflags, that says "this has not special
377 * permission function, use the fast case".
379 static inline int do_inode_permission(struct inode *inode, int mask)
381 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
382 if (likely(inode->i_op->permission))
383 return inode->i_op->permission(inode, mask);
385 /* This gets set once for the inode lifetime */
386 spin_lock(&inode->i_lock);
387 inode->i_opflags |= IOP_FASTPERM;
388 spin_unlock(&inode->i_lock);
390 return generic_permission(inode, mask);
394 * __inode_permission - Check for access rights to a given inode
395 * @inode: Inode to check permission on
396 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
398 * Check for read/write/execute permissions on an inode.
400 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
402 * This does not check for a read-only file system. You probably want
403 * inode_permission().
405 int __inode_permission(struct inode *inode, int mask)
409 if (unlikely(mask & MAY_WRITE)) {
411 * Nobody gets write access to an immutable file.
413 if (IS_IMMUTABLE(inode))
417 * Updating mtime will likely cause i_uid and i_gid to be
418 * written back improperly if their true value is unknown
421 if (HAS_UNMAPPED_ID(inode))
425 retval = do_inode_permission(inode, mask);
429 retval = devcgroup_inode_permission(inode, mask);
433 return security_inode_permission(inode, mask);
435 EXPORT_SYMBOL(__inode_permission);
438 * sb_permission - Check superblock-level permissions
439 * @sb: Superblock of inode to check permission on
440 * @inode: Inode to check permission on
441 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
443 * Separate out file-system wide checks from inode-specific permission checks.
445 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
447 if (unlikely(mask & MAY_WRITE)) {
448 umode_t mode = inode->i_mode;
450 /* Nobody gets write access to a read-only fs. */
451 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
468 int inode_permission(struct inode *inode, int mask)
472 retval = sb_permission(inode->i_sb, inode, mask);
475 return __inode_permission(inode, mask);
477 EXPORT_SYMBOL(inode_permission);
480 * path_get - get a reference to a path
481 * @path: path to get the reference to
483 * Given a path increment the reference count to the dentry and the vfsmount.
485 void path_get(const struct path *path)
490 EXPORT_SYMBOL(path_get);
493 * path_put - put a reference to a path
494 * @path: path to put the reference to
496 * Given a path decrement the reference count to the dentry and the vfsmount.
498 void path_put(const struct path *path)
503 EXPORT_SYMBOL(path_put);
505 #define EMBEDDED_LEVELS 2
510 struct inode *inode; /* path.dentry.d_inode */
515 int total_link_count;
518 struct delayed_call done;
521 } *stack, internal[EMBEDDED_LEVELS];
522 struct filename *name;
523 struct nameidata *saved;
524 struct inode *link_inode;
527 } __randomize_layout;
529 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
535 p->total_link_count = old ? old->total_link_count : 0;
537 current->nameidata = p;
540 static void restore_nameidata(void)
542 struct nameidata *now = current->nameidata, *old = now->saved;
544 current->nameidata = old;
546 old->total_link_count = now->total_link_count;
547 if (now->stack != now->internal)
551 static int __nd_alloc_stack(struct nameidata *nd)
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
566 memcpy(p, nd->internal, sizeof(nd->internal));
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
578 static bool path_connected(const struct path *path)
580 struct vfsmount *mnt = path->mnt;
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
586 return is_subdir(path->dentry, mnt->mnt_root);
589 static inline int nd_alloc_stack(struct nameidata *nd)
591 if (likely(nd->depth != EMBEDDED_LEVELS))
593 if (likely(nd->stack != nd->internal))
595 return __nd_alloc_stack(nd);
598 static void drop_links(struct nameidata *nd)
602 struct saved *last = nd->stack + i;
603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
608 static void terminate_walk(struct nameidata *nd)
611 if (!(nd->flags & LOOKUP_RCU)) {
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
629 /* path_put is needed afterwards regardless of success or failure */
630 static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
647 static bool legitimize_links(struct nameidata *nd)
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
662 * Path walking has 2 modes, rcu-walk and ref-walk (see
663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * Returns: 0 on success, -ECHILD on failure
677 * unlazy_walk attempts to legitimize the current nd->path and nd->root
679 * Must be called from rcu-walk context.
680 * Nothing should touch nameidata between unlazy_walk() failure and
683 static int unlazy_walk(struct nameidata *nd)
685 struct dentry *parent = nd->path.dentry;
687 BUG_ON(!(nd->flags & LOOKUP_RCU));
689 nd->flags &= ~LOOKUP_RCU;
690 if (unlikely(!legitimize_links(nd)))
692 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
694 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
695 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
699 BUG_ON(nd->inode != parent->d_inode);
704 nd->path.dentry = NULL;
706 if (!(nd->flags & LOOKUP_ROOT))
714 * unlazy_child - try to switch to ref-walk mode.
715 * @nd: nameidata pathwalk data
716 * @dentry: child of nd->path.dentry
717 * @seq: seq number to check dentry against
718 * Returns: 0 on success, -ECHILD on failure
720 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
721 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
722 * @nd. Must be called from rcu-walk context.
723 * Nothing should touch nameidata between unlazy_child() failure and
726 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
728 BUG_ON(!(nd->flags & LOOKUP_RCU));
730 nd->flags &= ~LOOKUP_RCU;
731 if (unlikely(!legitimize_links(nd)))
733 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
735 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
739 * We need to move both the parent and the dentry from the RCU domain
740 * to be properly refcounted. And the sequence number in the dentry
741 * validates *both* dentry counters, since we checked the sequence
742 * number of the parent after we got the child sequence number. So we
743 * know the parent must still be valid if the child sequence number is
745 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
747 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
753 * Sequence counts matched. Now make sure that the root is
754 * still valid and get it if required.
756 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
757 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
770 nd->path.dentry = NULL;
774 if (!(nd->flags & LOOKUP_ROOT))
779 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
781 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
782 return dentry->d_op->d_revalidate(dentry, flags);
788 * complete_walk - successful completion of path walk
789 * @nd: pointer nameidata
791 * If we had been in RCU mode, drop out of it and legitimize nd->path.
792 * Revalidate the final result, unless we'd already done that during
793 * the path walk or the filesystem doesn't ask for it. Return 0 on
794 * success, -error on failure. In case of failure caller does not
795 * need to drop nd->path.
797 static int complete_walk(struct nameidata *nd)
799 struct dentry *dentry = nd->path.dentry;
802 if (nd->flags & LOOKUP_RCU) {
803 if (!(nd->flags & LOOKUP_ROOT))
805 if (unlikely(unlazy_walk(nd)))
809 if (likely(!(nd->flags & LOOKUP_JUMPED)))
812 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
815 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
825 static void set_root(struct nameidata *nd)
827 struct fs_struct *fs = current->fs;
829 if (nd->flags & LOOKUP_RCU) {
833 seq = read_seqcount_begin(&fs->seq);
835 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
836 } while (read_seqcount_retry(&fs->seq, seq));
838 get_fs_root(fs, &nd->root);
842 static void path_put_conditional(struct path *path, struct nameidata *nd)
845 if (path->mnt != nd->path.mnt)
849 static inline void path_to_nameidata(const struct path *path,
850 struct nameidata *nd)
852 if (!(nd->flags & LOOKUP_RCU)) {
853 dput(nd->path.dentry);
854 if (nd->path.mnt != path->mnt)
855 mntput(nd->path.mnt);
857 nd->path.mnt = path->mnt;
858 nd->path.dentry = path->dentry;
861 static int nd_jump_root(struct nameidata *nd)
863 if (nd->flags & LOOKUP_RCU) {
867 nd->inode = d->d_inode;
868 nd->seq = nd->root_seq;
869 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
875 nd->inode = nd->path.dentry->d_inode;
877 nd->flags |= LOOKUP_JUMPED;
882 * Helper to directly jump to a known parsed path from ->get_link,
883 * caller must have taken a reference to path beforehand.
885 void nd_jump_link(struct path *path)
887 struct nameidata *nd = current->nameidata;
891 nd->inode = nd->path.dentry->d_inode;
892 nd->flags |= LOOKUP_JUMPED;
895 static inline void put_link(struct nameidata *nd)
897 struct saved *last = nd->stack + --nd->depth;
898 do_delayed_call(&last->done);
899 if (!(nd->flags & LOOKUP_RCU))
900 path_put(&last->link);
903 int sysctl_protected_symlinks __read_mostly = 0;
904 int sysctl_protected_hardlinks __read_mostly = 0;
907 * may_follow_link - Check symlink following for unsafe situations
908 * @nd: nameidata pathwalk data
910 * In the case of the sysctl_protected_symlinks sysctl being enabled,
911 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
912 * in a sticky world-writable directory. This is to protect privileged
913 * processes from failing races against path names that may change out
914 * from under them by way of other users creating malicious symlinks.
915 * It will permit symlinks to be followed only when outside a sticky
916 * world-writable directory, or when the uid of the symlink and follower
917 * match, or when the directory owner matches the symlink's owner.
919 * Returns 0 if following the symlink is allowed, -ve on error.
921 static inline int may_follow_link(struct nameidata *nd)
923 const struct inode *inode;
924 const struct inode *parent;
927 if (!sysctl_protected_symlinks)
930 /* Allowed if owner and follower match. */
931 inode = nd->link_inode;
932 if (uid_eq(current_cred()->fsuid, inode->i_uid))
935 /* Allowed if parent directory not sticky and world-writable. */
937 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
940 /* Allowed if parent directory and link owner match. */
941 puid = parent->i_uid;
942 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
945 if (nd->flags & LOOKUP_RCU)
948 audit_log_link_denied("follow_link", &nd->stack[0].link);
953 * safe_hardlink_source - Check for safe hardlink conditions
954 * @inode: the source inode to hardlink from
956 * Return false if at least one of the following conditions:
957 * - inode is not a regular file
959 * - inode is setgid and group-exec
960 * - access failure for read and write
962 * Otherwise returns true.
964 static bool safe_hardlink_source(struct inode *inode)
966 umode_t mode = inode->i_mode;
968 /* Special files should not get pinned to the filesystem. */
972 /* Setuid files should not get pinned to the filesystem. */
976 /* Executable setgid files should not get pinned to the filesystem. */
977 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
980 /* Hardlinking to unreadable or unwritable sources is dangerous. */
981 if (inode_permission(inode, MAY_READ | MAY_WRITE))
988 * may_linkat - Check permissions for creating a hardlink
989 * @link: the source to hardlink from
991 * Block hardlink when all of:
992 * - sysctl_protected_hardlinks enabled
993 * - fsuid does not match inode
994 * - hardlink source is unsafe (see safe_hardlink_source() above)
995 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
997 * Returns 0 if successful, -ve on error.
999 static int may_linkat(struct path *link)
1001 struct inode *inode;
1003 if (!sysctl_protected_hardlinks)
1006 inode = link->dentry->d_inode;
1008 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1009 * otherwise, it must be a safe source.
1011 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1014 audit_log_link_denied("linkat", link);
1018 static __always_inline
1019 const char *get_link(struct nameidata *nd)
1021 struct saved *last = nd->stack + nd->depth - 1;
1022 struct dentry *dentry = last->link.dentry;
1023 struct inode *inode = nd->link_inode;
1027 if (!(nd->flags & LOOKUP_RCU)) {
1028 touch_atime(&last->link);
1030 } else if (atime_needs_update_rcu(&last->link, inode)) {
1031 if (unlikely(unlazy_walk(nd)))
1032 return ERR_PTR(-ECHILD);
1033 touch_atime(&last->link);
1036 error = security_inode_follow_link(dentry, inode,
1037 nd->flags & LOOKUP_RCU);
1038 if (unlikely(error))
1039 return ERR_PTR(error);
1041 nd->last_type = LAST_BIND;
1042 res = inode->i_link;
1044 const char * (*get)(struct dentry *, struct inode *,
1045 struct delayed_call *);
1046 get = inode->i_op->get_link;
1047 if (nd->flags & LOOKUP_RCU) {
1048 res = get(NULL, inode, &last->done);
1049 if (res == ERR_PTR(-ECHILD)) {
1050 if (unlikely(unlazy_walk(nd)))
1051 return ERR_PTR(-ECHILD);
1052 res = get(dentry, inode, &last->done);
1055 res = get(dentry, inode, &last->done);
1057 if (IS_ERR_OR_NULL(res))
1063 if (unlikely(nd_jump_root(nd)))
1064 return ERR_PTR(-ECHILD);
1065 while (unlikely(*++res == '/'))
1074 * follow_up - Find the mountpoint of path's vfsmount
1076 * Given a path, find the mountpoint of its source file system.
1077 * Replace @path with the path of the mountpoint in the parent mount.
1080 * Return 1 if we went up a level and 0 if we were already at the
1083 int follow_up(struct path *path)
1085 struct mount *mnt = real_mount(path->mnt);
1086 struct mount *parent;
1087 struct dentry *mountpoint;
1089 read_seqlock_excl(&mount_lock);
1090 parent = mnt->mnt_parent;
1091 if (parent == mnt) {
1092 read_sequnlock_excl(&mount_lock);
1095 mntget(&parent->mnt);
1096 mountpoint = dget(mnt->mnt_mountpoint);
1097 read_sequnlock_excl(&mount_lock);
1099 path->dentry = mountpoint;
1101 path->mnt = &parent->mnt;
1104 EXPORT_SYMBOL(follow_up);
1107 * Perform an automount
1108 * - return -EISDIR to tell follow_managed() to stop and return the path we
1111 static int follow_automount(struct path *path, struct nameidata *nd,
1114 struct vfsmount *mnt;
1117 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1120 /* We don't want to mount if someone's just doing a stat -
1121 * unless they're stat'ing a directory and appended a '/' to
1124 * We do, however, want to mount if someone wants to open or
1125 * create a file of any type under the mountpoint, wants to
1126 * traverse through the mountpoint or wants to open the
1127 * mounted directory. Also, autofs may mark negative dentries
1128 * as being automount points. These will need the attentions
1129 * of the daemon to instantiate them before they can be used.
1131 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1132 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1133 path->dentry->d_inode)
1136 nd->total_link_count++;
1137 if (nd->total_link_count >= 40)
1140 mnt = path->dentry->d_op->d_automount(path);
1143 * The filesystem is allowed to return -EISDIR here to indicate
1144 * it doesn't want to automount. For instance, autofs would do
1145 * this so that its userspace daemon can mount on this dentry.
1147 * However, we can only permit this if it's a terminal point in
1148 * the path being looked up; if it wasn't then the remainder of
1149 * the path is inaccessible and we should say so.
1151 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1153 return PTR_ERR(mnt);
1156 if (!mnt) /* mount collision */
1159 if (!*need_mntput) {
1160 /* lock_mount() may release path->mnt on error */
1162 *need_mntput = true;
1164 err = finish_automount(mnt, path);
1168 /* Someone else made a mount here whilst we were busy */
1173 path->dentry = dget(mnt->mnt_root);
1182 * Handle a dentry that is managed in some way.
1183 * - Flagged for transit management (autofs)
1184 * - Flagged as mountpoint
1185 * - Flagged as automount point
1187 * This may only be called in refwalk mode.
1189 * Serialization is taken care of in namespace.c
1191 static int follow_managed(struct path *path, struct nameidata *nd)
1193 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1195 bool need_mntput = false;
1198 /* Given that we're not holding a lock here, we retain the value in a
1199 * local variable for each dentry as we look at it so that we don't see
1200 * the components of that value change under us */
1201 while (managed = READ_ONCE(path->dentry->d_flags),
1202 managed &= DCACHE_MANAGED_DENTRY,
1203 unlikely(managed != 0)) {
1204 /* Allow the filesystem to manage the transit without i_mutex
1206 if (managed & DCACHE_MANAGE_TRANSIT) {
1207 BUG_ON(!path->dentry->d_op);
1208 BUG_ON(!path->dentry->d_op->d_manage);
1209 ret = path->dentry->d_op->d_manage(path, false);
1214 /* Transit to a mounted filesystem. */
1215 if (managed & DCACHE_MOUNTED) {
1216 struct vfsmount *mounted = lookup_mnt(path);
1221 path->mnt = mounted;
1222 path->dentry = dget(mounted->mnt_root);
1227 /* Something is mounted on this dentry in another
1228 * namespace and/or whatever was mounted there in this
1229 * namespace got unmounted before lookup_mnt() could
1233 /* Handle an automount point */
1234 if (managed & DCACHE_NEED_AUTOMOUNT) {
1235 ret = follow_automount(path, nd, &need_mntput);
1241 /* We didn't change the current path point */
1245 if (need_mntput && path->mnt == mnt)
1247 if (ret == -EISDIR || !ret)
1250 nd->flags |= LOOKUP_JUMPED;
1251 if (unlikely(ret < 0))
1252 path_put_conditional(path, nd);
1256 int follow_down_one(struct path *path)
1258 struct vfsmount *mounted;
1260 mounted = lookup_mnt(path);
1264 path->mnt = mounted;
1265 path->dentry = dget(mounted->mnt_root);
1270 EXPORT_SYMBOL(follow_down_one);
1272 static inline int managed_dentry_rcu(const struct path *path)
1274 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1275 path->dentry->d_op->d_manage(path, true) : 0;
1279 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1280 * we meet a managed dentry that would need blocking.
1282 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1283 struct inode **inode, unsigned *seqp)
1286 struct mount *mounted;
1288 * Don't forget we might have a non-mountpoint managed dentry
1289 * that wants to block transit.
1291 switch (managed_dentry_rcu(path)) {
1301 if (!d_mountpoint(path->dentry))
1302 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1304 mounted = __lookup_mnt(path->mnt, path->dentry);
1307 path->mnt = &mounted->mnt;
1308 path->dentry = mounted->mnt.mnt_root;
1309 nd->flags |= LOOKUP_JUMPED;
1310 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1312 * Update the inode too. We don't need to re-check the
1313 * dentry sequence number here after this d_inode read,
1314 * because a mount-point is always pinned.
1316 *inode = path->dentry->d_inode;
1318 return !read_seqretry(&mount_lock, nd->m_seq) &&
1319 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1322 static int follow_dotdot_rcu(struct nameidata *nd)
1324 struct inode *inode = nd->inode;
1327 if (path_equal(&nd->path, &nd->root))
1329 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1330 struct dentry *old = nd->path.dentry;
1331 struct dentry *parent = old->d_parent;
1334 inode = parent->d_inode;
1335 seq = read_seqcount_begin(&parent->d_seq);
1336 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1338 nd->path.dentry = parent;
1340 if (unlikely(!path_connected(&nd->path)))
1344 struct mount *mnt = real_mount(nd->path.mnt);
1345 struct mount *mparent = mnt->mnt_parent;
1346 struct dentry *mountpoint = mnt->mnt_mountpoint;
1347 struct inode *inode2 = mountpoint->d_inode;
1348 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1349 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1351 if (&mparent->mnt == nd->path.mnt)
1353 /* we know that mountpoint was pinned */
1354 nd->path.dentry = mountpoint;
1355 nd->path.mnt = &mparent->mnt;
1360 while (unlikely(d_mountpoint(nd->path.dentry))) {
1361 struct mount *mounted;
1362 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1363 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1367 nd->path.mnt = &mounted->mnt;
1368 nd->path.dentry = mounted->mnt.mnt_root;
1369 inode = nd->path.dentry->d_inode;
1370 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1377 * Follow down to the covering mount currently visible to userspace. At each
1378 * point, the filesystem owning that dentry may be queried as to whether the
1379 * caller is permitted to proceed or not.
1381 int follow_down(struct path *path)
1386 while (managed = READ_ONCE(path->dentry->d_flags),
1387 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1388 /* Allow the filesystem to manage the transit without i_mutex
1391 * We indicate to the filesystem if someone is trying to mount
1392 * something here. This gives autofs the chance to deny anyone
1393 * other than its daemon the right to mount on its
1396 * The filesystem may sleep at this point.
1398 if (managed & DCACHE_MANAGE_TRANSIT) {
1399 BUG_ON(!path->dentry->d_op);
1400 BUG_ON(!path->dentry->d_op->d_manage);
1401 ret = path->dentry->d_op->d_manage(path, false);
1403 return ret == -EISDIR ? 0 : ret;
1406 /* Transit to a mounted filesystem. */
1407 if (managed & DCACHE_MOUNTED) {
1408 struct vfsmount *mounted = lookup_mnt(path);
1413 path->mnt = mounted;
1414 path->dentry = dget(mounted->mnt_root);
1418 /* Don't handle automount points here */
1423 EXPORT_SYMBOL(follow_down);
1426 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1428 static void follow_mount(struct path *path)
1430 while (d_mountpoint(path->dentry)) {
1431 struct vfsmount *mounted = lookup_mnt(path);
1436 path->mnt = mounted;
1437 path->dentry = dget(mounted->mnt_root);
1441 static int path_parent_directory(struct path *path)
1443 struct dentry *old = path->dentry;
1444 /* rare case of legitimate dget_parent()... */
1445 path->dentry = dget_parent(path->dentry);
1447 if (unlikely(!path_connected(path)))
1452 static int follow_dotdot(struct nameidata *nd)
1455 if (nd->path.dentry == nd->root.dentry &&
1456 nd->path.mnt == nd->root.mnt) {
1459 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1460 int ret = path_parent_directory(&nd->path);
1465 if (!follow_up(&nd->path))
1468 follow_mount(&nd->path);
1469 nd->inode = nd->path.dentry->d_inode;
1474 * This looks up the name in dcache and possibly revalidates the found dentry.
1475 * NULL is returned if the dentry does not exist in the cache.
1477 static struct dentry *lookup_dcache(const struct qstr *name,
1481 struct dentry *dentry = d_lookup(dir, name);
1483 int error = d_revalidate(dentry, flags);
1484 if (unlikely(error <= 0)) {
1486 d_invalidate(dentry);
1488 return ERR_PTR(error);
1495 * Call i_op->lookup on the dentry. The dentry must be negative and
1498 * dir->d_inode->i_mutex must be held
1500 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1505 /* Don't create child dentry for a dead directory. */
1506 if (unlikely(IS_DEADDIR(dir))) {
1508 return ERR_PTR(-ENOENT);
1511 old = dir->i_op->lookup(dir, dentry, flags);
1512 if (unlikely(old)) {
1519 static struct dentry *__lookup_hash(const struct qstr *name,
1520 struct dentry *base, unsigned int flags)
1522 struct dentry *dentry = lookup_dcache(name, base, flags);
1527 dentry = d_alloc(base, name);
1528 if (unlikely(!dentry))
1529 return ERR_PTR(-ENOMEM);
1531 return lookup_real(base->d_inode, dentry, flags);
1534 static int lookup_fast(struct nameidata *nd,
1535 struct path *path, struct inode **inode,
1538 struct vfsmount *mnt = nd->path.mnt;
1539 struct dentry *dentry, *parent = nd->path.dentry;
1544 * Rename seqlock is not required here because in the off chance
1545 * of a false negative due to a concurrent rename, the caller is
1546 * going to fall back to non-racy lookup.
1548 if (nd->flags & LOOKUP_RCU) {
1551 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1552 if (unlikely(!dentry)) {
1553 if (unlazy_walk(nd))
1559 * This sequence count validates that the inode matches
1560 * the dentry name information from lookup.
1562 *inode = d_backing_inode(dentry);
1563 negative = d_is_negative(dentry);
1564 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1568 * This sequence count validates that the parent had no
1569 * changes while we did the lookup of the dentry above.
1571 * The memory barrier in read_seqcount_begin of child is
1572 * enough, we can use __read_seqcount_retry here.
1574 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1578 status = d_revalidate(dentry, nd->flags);
1579 if (likely(status > 0)) {
1581 * Note: do negative dentry check after revalidation in
1582 * case that drops it.
1584 if (unlikely(negative))
1587 path->dentry = dentry;
1588 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1591 if (unlazy_child(nd, dentry, seq))
1593 if (unlikely(status == -ECHILD))
1594 /* we'd been told to redo it in non-rcu mode */
1595 status = d_revalidate(dentry, nd->flags);
1597 dentry = __d_lookup(parent, &nd->last);
1598 if (unlikely(!dentry))
1600 status = d_revalidate(dentry, nd->flags);
1602 if (unlikely(status <= 0)) {
1604 d_invalidate(dentry);
1608 if (unlikely(d_is_negative(dentry))) {
1614 path->dentry = dentry;
1615 err = follow_managed(path, nd);
1616 if (likely(err > 0))
1617 *inode = d_backing_inode(path->dentry);
1621 /* Fast lookup failed, do it the slow way */
1622 static struct dentry *lookup_slow(const struct qstr *name,
1626 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1627 struct inode *inode = dir->d_inode;
1628 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1630 inode_lock_shared(inode);
1631 /* Don't go there if it's already dead */
1632 if (unlikely(IS_DEADDIR(inode)))
1635 dentry = d_alloc_parallel(dir, name, &wq);
1638 if (unlikely(!d_in_lookup(dentry))) {
1639 if (!(flags & LOOKUP_NO_REVAL)) {
1640 int error = d_revalidate(dentry, flags);
1641 if (unlikely(error <= 0)) {
1643 d_invalidate(dentry);
1648 dentry = ERR_PTR(error);
1652 old = inode->i_op->lookup(inode, dentry, flags);
1653 d_lookup_done(dentry);
1654 if (unlikely(old)) {
1660 inode_unlock_shared(inode);
1664 static inline int may_lookup(struct nameidata *nd)
1666 if (nd->flags & LOOKUP_RCU) {
1667 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1670 if (unlazy_walk(nd))
1673 return inode_permission(nd->inode, MAY_EXEC);
1676 static inline int handle_dots(struct nameidata *nd, int type)
1678 if (type == LAST_DOTDOT) {
1681 if (nd->flags & LOOKUP_RCU) {
1682 return follow_dotdot_rcu(nd);
1684 return follow_dotdot(nd);
1689 static int pick_link(struct nameidata *nd, struct path *link,
1690 struct inode *inode, unsigned seq)
1694 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1695 path_to_nameidata(link, nd);
1698 if (!(nd->flags & LOOKUP_RCU)) {
1699 if (link->mnt == nd->path.mnt)
1702 error = nd_alloc_stack(nd);
1703 if (unlikely(error)) {
1704 if (error == -ECHILD) {
1705 if (unlikely(!legitimize_path(nd, link, seq))) {
1708 nd->flags &= ~LOOKUP_RCU;
1709 nd->path.mnt = NULL;
1710 nd->path.dentry = NULL;
1711 if (!(nd->flags & LOOKUP_ROOT))
1712 nd->root.mnt = NULL;
1714 } else if (likely(unlazy_walk(nd)) == 0)
1715 error = nd_alloc_stack(nd);
1723 last = nd->stack + nd->depth++;
1725 clear_delayed_call(&last->done);
1726 nd->link_inode = inode;
1731 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1734 * Do we need to follow links? We _really_ want to be able
1735 * to do this check without having to look at inode->i_op,
1736 * so we keep a cache of "no, this doesn't need follow_link"
1737 * for the common case.
1739 static inline int step_into(struct nameidata *nd, struct path *path,
1740 int flags, struct inode *inode, unsigned seq)
1742 if (!(flags & WALK_MORE) && nd->depth)
1744 if (likely(!d_is_symlink(path->dentry)) ||
1745 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1746 /* not a symlink or should not follow */
1747 path_to_nameidata(path, nd);
1752 /* make sure that d_is_symlink above matches inode */
1753 if (nd->flags & LOOKUP_RCU) {
1754 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1757 return pick_link(nd, path, inode, seq);
1760 static int walk_component(struct nameidata *nd, int flags)
1763 struct inode *inode;
1767 * "." and ".." are special - ".." especially so because it has
1768 * to be able to know about the current root directory and
1769 * parent relationships.
1771 if (unlikely(nd->last_type != LAST_NORM)) {
1772 err = handle_dots(nd, nd->last_type);
1773 if (!(flags & WALK_MORE) && nd->depth)
1777 err = lookup_fast(nd, &path, &inode, &seq);
1778 if (unlikely(err <= 0)) {
1781 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1783 if (IS_ERR(path.dentry))
1784 return PTR_ERR(path.dentry);
1786 path.mnt = nd->path.mnt;
1787 err = follow_managed(&path, nd);
1788 if (unlikely(err < 0))
1791 if (unlikely(d_is_negative(path.dentry))) {
1792 path_to_nameidata(&path, nd);
1796 seq = 0; /* we are already out of RCU mode */
1797 inode = d_backing_inode(path.dentry);
1800 return step_into(nd, &path, flags, inode, seq);
1804 * We can do the critical dentry name comparison and hashing
1805 * operations one word at a time, but we are limited to:
1807 * - Architectures with fast unaligned word accesses. We could
1808 * do a "get_unaligned()" if this helps and is sufficiently
1811 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1812 * do not trap on the (extremely unlikely) case of a page
1813 * crossing operation.
1815 * - Furthermore, we need an efficient 64-bit compile for the
1816 * 64-bit case in order to generate the "number of bytes in
1817 * the final mask". Again, that could be replaced with a
1818 * efficient population count instruction or similar.
1820 #ifdef CONFIG_DCACHE_WORD_ACCESS
1822 #include <asm/word-at-a-time.h>
1826 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1828 #elif defined(CONFIG_64BIT)
1830 * Register pressure in the mixing function is an issue, particularly
1831 * on 32-bit x86, but almost any function requires one state value and
1832 * one temporary. Instead, use a function designed for two state values
1833 * and no temporaries.
1835 * This function cannot create a collision in only two iterations, so
1836 * we have two iterations to achieve avalanche. In those two iterations,
1837 * we have six layers of mixing, which is enough to spread one bit's
1838 * influence out to 2^6 = 64 state bits.
1840 * Rotate constants are scored by considering either 64 one-bit input
1841 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1842 * probability of that delta causing a change to each of the 128 output
1843 * bits, using a sample of random initial states.
1845 * The Shannon entropy of the computed probabilities is then summed
1846 * to produce a score. Ideally, any input change has a 50% chance of
1847 * toggling any given output bit.
1849 * Mixing scores (in bits) for (12,45):
1850 * Input delta: 1-bit 2-bit
1851 * 1 round: 713.3 42542.6
1852 * 2 rounds: 2753.7 140389.8
1853 * 3 rounds: 5954.1 233458.2
1854 * 4 rounds: 7862.6 256672.2
1855 * Perfect: 8192 258048
1856 * (64*128) (64*63/2 * 128)
1858 #define HASH_MIX(x, y, a) \
1860 y ^= x, x = rol64(x,12),\
1861 x += y, y = rol64(y,45),\
1865 * Fold two longs into one 32-bit hash value. This must be fast, but
1866 * latency isn't quite as critical, as there is a fair bit of additional
1867 * work done before the hash value is used.
1869 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1871 y ^= x * GOLDEN_RATIO_64;
1872 y *= GOLDEN_RATIO_64;
1876 #else /* 32-bit case */
1879 * Mixing scores (in bits) for (7,20):
1880 * Input delta: 1-bit 2-bit
1881 * 1 round: 330.3 9201.6
1882 * 2 rounds: 1246.4 25475.4
1883 * 3 rounds: 1907.1 31295.1
1884 * 4 rounds: 2042.3 31718.6
1885 * Perfect: 2048 31744
1886 * (32*64) (32*31/2 * 64)
1888 #define HASH_MIX(x, y, a) \
1890 y ^= x, x = rol32(x, 7),\
1891 x += y, y = rol32(y,20),\
1894 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1896 /* Use arch-optimized multiply if one exists */
1897 return __hash_32(y ^ __hash_32(x));
1903 * Return the hash of a string of known length. This is carfully
1904 * designed to match hash_name(), which is the more critical function.
1905 * In particular, we must end by hashing a final word containing 0..7
1906 * payload bytes, to match the way that hash_name() iterates until it
1907 * finds the delimiter after the name.
1909 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1911 unsigned long a, x = 0, y = (unsigned long)salt;
1916 a = load_unaligned_zeropad(name);
1917 if (len < sizeof(unsigned long))
1920 name += sizeof(unsigned long);
1921 len -= sizeof(unsigned long);
1923 x ^= a & bytemask_from_count(len);
1925 return fold_hash(x, y);
1927 EXPORT_SYMBOL(full_name_hash);
1929 /* Return the "hash_len" (hash and length) of a null-terminated string */
1930 u64 hashlen_string(const void *salt, const char *name)
1932 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1933 unsigned long adata, mask, len;
1934 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1941 len += sizeof(unsigned long);
1943 a = load_unaligned_zeropad(name+len);
1944 } while (!has_zero(a, &adata, &constants));
1946 adata = prep_zero_mask(a, adata, &constants);
1947 mask = create_zero_mask(adata);
1948 x ^= a & zero_bytemask(mask);
1950 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1952 EXPORT_SYMBOL(hashlen_string);
1955 * Calculate the length and hash of the path component, and
1956 * return the "hash_len" as the result.
1958 static inline u64 hash_name(const void *salt, const char *name)
1960 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1961 unsigned long adata, bdata, mask, len;
1962 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1969 len += sizeof(unsigned long);
1971 a = load_unaligned_zeropad(name+len);
1972 b = a ^ REPEAT_BYTE('/');
1973 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1975 adata = prep_zero_mask(a, adata, &constants);
1976 bdata = prep_zero_mask(b, bdata, &constants);
1977 mask = create_zero_mask(adata | bdata);
1978 x ^= a & zero_bytemask(mask);
1980 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1983 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1985 /* Return the hash of a string of known length */
1986 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1988 unsigned long hash = init_name_hash(salt);
1990 hash = partial_name_hash((unsigned char)*name++, hash);
1991 return end_name_hash(hash);
1993 EXPORT_SYMBOL(full_name_hash);
1995 /* Return the "hash_len" (hash and length) of a null-terminated string */
1996 u64 hashlen_string(const void *salt, const char *name)
1998 unsigned long hash = init_name_hash(salt);
1999 unsigned long len = 0, c;
2001 c = (unsigned char)*name;
2004 hash = partial_name_hash(c, hash);
2005 c = (unsigned char)name[len];
2007 return hashlen_create(end_name_hash(hash), len);
2009 EXPORT_SYMBOL(hashlen_string);
2012 * We know there's a real path component here of at least
2015 static inline u64 hash_name(const void *salt, const char *name)
2017 unsigned long hash = init_name_hash(salt);
2018 unsigned long len = 0, c;
2020 c = (unsigned char)*name;
2023 hash = partial_name_hash(c, hash);
2024 c = (unsigned char)name[len];
2025 } while (c && c != '/');
2026 return hashlen_create(end_name_hash(hash), len);
2033 * This is the basic name resolution function, turning a pathname into
2034 * the final dentry. We expect 'base' to be positive and a directory.
2036 * Returns 0 and nd will have valid dentry and mnt on success.
2037 * Returns error and drops reference to input namei data on failure.
2039 static int link_path_walk(const char *name, struct nameidata *nd)
2048 /* At this point we know we have a real path component. */
2053 err = may_lookup(nd);
2057 hash_len = hash_name(nd->path.dentry, name);
2060 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2062 if (name[1] == '.') {
2064 nd->flags |= LOOKUP_JUMPED;
2070 if (likely(type == LAST_NORM)) {
2071 struct dentry *parent = nd->path.dentry;
2072 nd->flags &= ~LOOKUP_JUMPED;
2073 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2074 struct qstr this = { { .hash_len = hash_len }, .name = name };
2075 err = parent->d_op->d_hash(parent, &this);
2078 hash_len = this.hash_len;
2083 nd->last.hash_len = hash_len;
2084 nd->last.name = name;
2085 nd->last_type = type;
2087 name += hashlen_len(hash_len);
2091 * If it wasn't NUL, we know it was '/'. Skip that
2092 * slash, and continue until no more slashes.
2096 } while (unlikely(*name == '/'));
2097 if (unlikely(!*name)) {
2099 /* pathname body, done */
2102 name = nd->stack[nd->depth - 1].name;
2103 /* trailing symlink, done */
2106 /* last component of nested symlink */
2107 err = walk_component(nd, WALK_FOLLOW);
2109 /* not the last component */
2110 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2116 const char *s = get_link(nd);
2125 nd->stack[nd->depth - 1].name = name;
2130 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2131 if (nd->flags & LOOKUP_RCU) {
2132 if (unlazy_walk(nd))
2140 static const char *path_init(struct nameidata *nd, unsigned flags)
2142 const char *s = nd->name->name;
2145 flags &= ~LOOKUP_RCU;
2147 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2148 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2150 if (flags & LOOKUP_ROOT) {
2151 struct dentry *root = nd->root.dentry;
2152 struct inode *inode = root->d_inode;
2153 if (*s && unlikely(!d_can_lookup(root)))
2154 return ERR_PTR(-ENOTDIR);
2155 nd->path = nd->root;
2157 if (flags & LOOKUP_RCU) {
2159 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2160 nd->root_seq = nd->seq;
2161 nd->m_seq = read_seqbegin(&mount_lock);
2163 path_get(&nd->path);
2168 nd->root.mnt = NULL;
2169 nd->path.mnt = NULL;
2170 nd->path.dentry = NULL;
2172 nd->m_seq = read_seqbegin(&mount_lock);
2174 if (flags & LOOKUP_RCU)
2177 if (likely(!nd_jump_root(nd)))
2179 nd->root.mnt = NULL;
2181 return ERR_PTR(-ECHILD);
2182 } else if (nd->dfd == AT_FDCWD) {
2183 if (flags & LOOKUP_RCU) {
2184 struct fs_struct *fs = current->fs;
2190 seq = read_seqcount_begin(&fs->seq);
2192 nd->inode = nd->path.dentry->d_inode;
2193 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2194 } while (read_seqcount_retry(&fs->seq, seq));
2196 get_fs_pwd(current->fs, &nd->path);
2197 nd->inode = nd->path.dentry->d_inode;
2201 /* Caller must check execute permissions on the starting path component */
2202 struct fd f = fdget_raw(nd->dfd);
2203 struct dentry *dentry;
2206 return ERR_PTR(-EBADF);
2208 dentry = f.file->f_path.dentry;
2211 if (!d_can_lookup(dentry)) {
2213 return ERR_PTR(-ENOTDIR);
2217 nd->path = f.file->f_path;
2218 if (flags & LOOKUP_RCU) {
2220 nd->inode = nd->path.dentry->d_inode;
2221 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2223 path_get(&nd->path);
2224 nd->inode = nd->path.dentry->d_inode;
2231 static const char *trailing_symlink(struct nameidata *nd)
2234 int error = may_follow_link(nd);
2235 if (unlikely(error))
2236 return ERR_PTR(error);
2237 nd->flags |= LOOKUP_PARENT;
2238 nd->stack[0].name = NULL;
2243 static inline int lookup_last(struct nameidata *nd)
2245 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2246 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2248 nd->flags &= ~LOOKUP_PARENT;
2249 return walk_component(nd, 0);
2252 static int handle_lookup_down(struct nameidata *nd)
2254 struct path path = nd->path;
2255 struct inode *inode = nd->inode;
2256 unsigned seq = nd->seq;
2259 if (nd->flags & LOOKUP_RCU) {
2261 * don't bother with unlazy_walk on failure - we are
2262 * at the very beginning of walk, so we lose nothing
2263 * if we simply redo everything in non-RCU mode
2265 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2269 err = follow_managed(&path, nd);
2270 if (unlikely(err < 0))
2272 inode = d_backing_inode(path.dentry);
2275 path_to_nameidata(&path, nd);
2281 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2282 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2284 const char *s = path_init(nd, flags);
2290 if (unlikely(flags & LOOKUP_DOWN)) {
2291 err = handle_lookup_down(nd);
2292 if (unlikely(err < 0)) {
2298 while (!(err = link_path_walk(s, nd))
2299 && ((err = lookup_last(nd)) > 0)) {
2300 s = trailing_symlink(nd);
2307 err = complete_walk(nd);
2309 if (!err && nd->flags & LOOKUP_DIRECTORY)
2310 if (!d_can_lookup(nd->path.dentry))
2314 nd->path.mnt = NULL;
2315 nd->path.dentry = NULL;
2321 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2322 struct path *path, struct path *root)
2325 struct nameidata nd;
2327 return PTR_ERR(name);
2328 if (unlikely(root)) {
2330 flags |= LOOKUP_ROOT;
2332 set_nameidata(&nd, dfd, name);
2333 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2334 if (unlikely(retval == -ECHILD))
2335 retval = path_lookupat(&nd, flags, path);
2336 if (unlikely(retval == -ESTALE))
2337 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2339 if (likely(!retval))
2340 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2341 restore_nameidata();
2346 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2347 static int path_parentat(struct nameidata *nd, unsigned flags,
2348 struct path *parent)
2350 const char *s = path_init(nd, flags);
2354 err = link_path_walk(s, nd);
2356 err = complete_walk(nd);
2359 nd->path.mnt = NULL;
2360 nd->path.dentry = NULL;
2366 static struct filename *filename_parentat(int dfd, struct filename *name,
2367 unsigned int flags, struct path *parent,
2368 struct qstr *last, int *type)
2371 struct nameidata nd;
2375 set_nameidata(&nd, dfd, name);
2376 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2377 if (unlikely(retval == -ECHILD))
2378 retval = path_parentat(&nd, flags, parent);
2379 if (unlikely(retval == -ESTALE))
2380 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2381 if (likely(!retval)) {
2383 *type = nd.last_type;
2384 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2387 name = ERR_PTR(retval);
2389 restore_nameidata();
2393 /* does lookup, returns the object with parent locked */
2394 struct dentry *kern_path_locked(const char *name, struct path *path)
2396 struct filename *filename;
2401 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2403 if (IS_ERR(filename))
2404 return ERR_CAST(filename);
2405 if (unlikely(type != LAST_NORM)) {
2408 return ERR_PTR(-EINVAL);
2410 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2411 d = __lookup_hash(&last, path->dentry, 0);
2413 inode_unlock(path->dentry->d_inode);
2420 int kern_path(const char *name, unsigned int flags, struct path *path)
2422 return filename_lookup(AT_FDCWD, getname_kernel(name),
2425 EXPORT_SYMBOL(kern_path);
2428 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2429 * @dentry: pointer to dentry of the base directory
2430 * @mnt: pointer to vfs mount of the base directory
2431 * @name: pointer to file name
2432 * @flags: lookup flags
2433 * @path: pointer to struct path to fill
2435 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2436 const char *name, unsigned int flags,
2439 struct path root = {.mnt = mnt, .dentry = dentry};
2440 /* the first argument of filename_lookup() is ignored with root */
2441 return filename_lookup(AT_FDCWD, getname_kernel(name),
2442 flags , path, &root);
2444 EXPORT_SYMBOL(vfs_path_lookup);
2447 * lookup_one_len - filesystem helper to lookup single pathname component
2448 * @name: pathname component to lookup
2449 * @base: base directory to lookup from
2450 * @len: maximum length @len should be interpreted to
2452 * Note that this routine is purely a helper for filesystem usage and should
2453 * not be called by generic code.
2455 * The caller must hold base->i_mutex.
2457 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2463 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2467 this.hash = full_name_hash(base, name, len);
2469 return ERR_PTR(-EACCES);
2471 if (unlikely(name[0] == '.')) {
2472 if (len < 2 || (len == 2 && name[1] == '.'))
2473 return ERR_PTR(-EACCES);
2477 c = *(const unsigned char *)name++;
2478 if (c == '/' || c == '\0')
2479 return ERR_PTR(-EACCES);
2482 * See if the low-level filesystem might want
2483 * to use its own hash..
2485 if (base->d_flags & DCACHE_OP_HASH) {
2486 int err = base->d_op->d_hash(base, &this);
2488 return ERR_PTR(err);
2491 err = inode_permission(base->d_inode, MAY_EXEC);
2493 return ERR_PTR(err);
2495 return __lookup_hash(&this, base, 0);
2497 EXPORT_SYMBOL(lookup_one_len);
2500 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2501 * @name: pathname component to lookup
2502 * @base: base directory to lookup from
2503 * @len: maximum length @len should be interpreted to
2505 * Note that this routine is purely a helper for filesystem usage and should
2506 * not be called by generic code.
2508 * Unlike lookup_one_len, it should be called without the parent
2509 * i_mutex held, and will take the i_mutex itself if necessary.
2511 struct dentry *lookup_one_len_unlocked(const char *name,
2512 struct dentry *base, int len)
2521 this.hash = full_name_hash(base, name, len);
2523 return ERR_PTR(-EACCES);
2525 if (unlikely(name[0] == '.')) {
2526 if (len < 2 || (len == 2 && name[1] == '.'))
2527 return ERR_PTR(-EACCES);
2531 c = *(const unsigned char *)name++;
2532 if (c == '/' || c == '\0')
2533 return ERR_PTR(-EACCES);
2536 * See if the low-level filesystem might want
2537 * to use its own hash..
2539 if (base->d_flags & DCACHE_OP_HASH) {
2540 int err = base->d_op->d_hash(base, &this);
2542 return ERR_PTR(err);
2545 err = inode_permission(base->d_inode, MAY_EXEC);
2547 return ERR_PTR(err);
2549 ret = lookup_dcache(&this, base, 0);
2551 ret = lookup_slow(&this, base, 0);
2554 EXPORT_SYMBOL(lookup_one_len_unlocked);
2556 #ifdef CONFIG_UNIX98_PTYS
2557 int path_pts(struct path *path)
2559 /* Find something mounted on "pts" in the same directory as
2562 struct dentry *child, *parent;
2566 ret = path_parent_directory(path);
2570 parent = path->dentry;
2573 child = d_hash_and_lookup(parent, &this);
2577 path->dentry = child;
2584 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2585 struct path *path, int *empty)
2587 return filename_lookup(dfd, getname_flags(name, flags, empty),
2590 EXPORT_SYMBOL(user_path_at_empty);
2593 * mountpoint_last - look up last component for umount
2594 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2596 * This is a special lookup_last function just for umount. In this case, we
2597 * need to resolve the path without doing any revalidation.
2599 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2600 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2601 * in almost all cases, this lookup will be served out of the dcache. The only
2602 * cases where it won't are if nd->last refers to a symlink or the path is
2603 * bogus and it doesn't exist.
2606 * -error: if there was an error during lookup. This includes -ENOENT if the
2607 * lookup found a negative dentry.
2609 * 0: if we successfully resolved nd->last and found it to not to be a
2610 * symlink that needs to be followed.
2612 * 1: if we successfully resolved nd->last and found it to be a symlink
2613 * that needs to be followed.
2616 mountpoint_last(struct nameidata *nd)
2619 struct dentry *dir = nd->path.dentry;
2622 /* If we're in rcuwalk, drop out of it to handle last component */
2623 if (nd->flags & LOOKUP_RCU) {
2624 if (unlazy_walk(nd))
2628 nd->flags &= ~LOOKUP_PARENT;
2630 if (unlikely(nd->last_type != LAST_NORM)) {
2631 error = handle_dots(nd, nd->last_type);
2634 path.dentry = dget(nd->path.dentry);
2636 path.dentry = d_lookup(dir, &nd->last);
2639 * No cached dentry. Mounted dentries are pinned in the
2640 * cache, so that means that this dentry is probably
2641 * a symlink or the path doesn't actually point
2642 * to a mounted dentry.
2644 path.dentry = lookup_slow(&nd->last, dir,
2645 nd->flags | LOOKUP_NO_REVAL);
2646 if (IS_ERR(path.dentry))
2647 return PTR_ERR(path.dentry);
2650 if (d_is_negative(path.dentry)) {
2654 path.mnt = nd->path.mnt;
2655 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2659 * path_mountpoint - look up a path to be umounted
2660 * @nd: lookup context
2661 * @flags: lookup flags
2662 * @path: pointer to container for result
2664 * Look up the given name, but don't attempt to revalidate the last component.
2665 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2668 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2670 const char *s = path_init(nd, flags);
2674 while (!(err = link_path_walk(s, nd)) &&
2675 (err = mountpoint_last(nd)) > 0) {
2676 s = trailing_symlink(nd);
2684 nd->path.mnt = NULL;
2685 nd->path.dentry = NULL;
2693 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2696 struct nameidata nd;
2699 return PTR_ERR(name);
2700 set_nameidata(&nd, dfd, name);
2701 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2702 if (unlikely(error == -ECHILD))
2703 error = path_mountpoint(&nd, flags, path);
2704 if (unlikely(error == -ESTALE))
2705 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2707 audit_inode(name, path->dentry, 0);
2708 restore_nameidata();
2714 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2715 * @dfd: directory file descriptor
2716 * @name: pathname from userland
2717 * @flags: lookup flags
2718 * @path: pointer to container to hold result
2720 * A umount is a special case for path walking. We're not actually interested
2721 * in the inode in this situation, and ESTALE errors can be a problem. We
2722 * simply want track down the dentry and vfsmount attached at the mountpoint
2723 * and avoid revalidating the last component.
2725 * Returns 0 and populates "path" on success.
2728 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2731 return filename_mountpoint(dfd, getname(name), path, flags);
2735 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2738 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2740 EXPORT_SYMBOL(kern_path_mountpoint);
2742 int __check_sticky(struct inode *dir, struct inode *inode)
2744 kuid_t fsuid = current_fsuid();
2746 if (uid_eq(inode->i_uid, fsuid))
2748 if (uid_eq(dir->i_uid, fsuid))
2750 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2752 EXPORT_SYMBOL(__check_sticky);
2755 * Check whether we can remove a link victim from directory dir, check
2756 * whether the type of victim is right.
2757 * 1. We can't do it if dir is read-only (done in permission())
2758 * 2. We should have write and exec permissions on dir
2759 * 3. We can't remove anything from append-only dir
2760 * 4. We can't do anything with immutable dir (done in permission())
2761 * 5. If the sticky bit on dir is set we should either
2762 * a. be owner of dir, or
2763 * b. be owner of victim, or
2764 * c. have CAP_FOWNER capability
2765 * 6. If the victim is append-only or immutable we can't do antyhing with
2766 * links pointing to it.
2767 * 7. If the victim has an unknown uid or gid we can't change the inode.
2768 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2769 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2770 * 10. We can't remove a root or mountpoint.
2771 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2772 * nfs_async_unlink().
2774 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2776 struct inode *inode = d_backing_inode(victim);
2779 if (d_is_negative(victim))
2783 BUG_ON(victim->d_parent->d_inode != dir);
2784 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2786 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2792 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2793 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2796 if (!d_is_dir(victim))
2798 if (IS_ROOT(victim))
2800 } else if (d_is_dir(victim))
2802 if (IS_DEADDIR(dir))
2804 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2809 /* Check whether we can create an object with dentry child in directory
2811 * 1. We can't do it if child already exists (open has special treatment for
2812 * this case, but since we are inlined it's OK)
2813 * 2. We can't do it if dir is read-only (done in permission())
2814 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2815 * 4. We should have write and exec permissions on dir
2816 * 5. We can't do it if dir is immutable (done in permission())
2818 static inline int may_create(struct inode *dir, struct dentry *child)
2820 struct user_namespace *s_user_ns;
2821 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2824 if (IS_DEADDIR(dir))
2826 s_user_ns = dir->i_sb->s_user_ns;
2827 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2828 !kgid_has_mapping(s_user_ns, current_fsgid()))
2830 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2834 * p1 and p2 should be directories on the same fs.
2836 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2841 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2845 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2847 p = d_ancestor(p2, p1);
2849 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2850 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2854 p = d_ancestor(p1, p2);
2856 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2857 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2861 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2862 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2865 EXPORT_SYMBOL(lock_rename);
2867 void unlock_rename(struct dentry *p1, struct dentry *p2)
2869 inode_unlock(p1->d_inode);
2871 inode_unlock(p2->d_inode);
2872 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2875 EXPORT_SYMBOL(unlock_rename);
2877 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2880 int error = may_create(dir, dentry);
2884 if (!dir->i_op->create)
2885 return -EACCES; /* shouldn't it be ENOSYS? */
2888 error = security_inode_create(dir, dentry, mode);
2891 error = dir->i_op->create(dir, dentry, mode, want_excl);
2893 fsnotify_create(dir, dentry);
2896 EXPORT_SYMBOL(vfs_create);
2898 bool may_open_dev(const struct path *path)
2900 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2901 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2904 static int may_open(const struct path *path, int acc_mode, int flag)
2906 struct dentry *dentry = path->dentry;
2907 struct inode *inode = dentry->d_inode;
2913 switch (inode->i_mode & S_IFMT) {
2917 if (acc_mode & MAY_WRITE)
2922 if (!may_open_dev(path))
2931 error = inode_permission(inode, MAY_OPEN | acc_mode);
2936 * An append-only file must be opened in append mode for writing.
2938 if (IS_APPEND(inode)) {
2939 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2945 /* O_NOATIME can only be set by the owner or superuser */
2946 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2952 static int handle_truncate(struct file *filp)
2954 const struct path *path = &filp->f_path;
2955 struct inode *inode = path->dentry->d_inode;
2956 int error = get_write_access(inode);
2960 * Refuse to truncate files with mandatory locks held on them.
2962 error = locks_verify_locked(filp);
2964 error = security_path_truncate(path);
2966 error = do_truncate(path->dentry, 0,
2967 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2970 put_write_access(inode);
2974 static inline int open_to_namei_flags(int flag)
2976 if ((flag & O_ACCMODE) == 3)
2981 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2983 struct user_namespace *s_user_ns;
2984 int error = security_path_mknod(dir, dentry, mode, 0);
2988 s_user_ns = dir->dentry->d_sb->s_user_ns;
2989 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2990 !kgid_has_mapping(s_user_ns, current_fsgid()))
2993 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2997 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3001 * Attempt to atomically look up, create and open a file from a negative
3004 * Returns 0 if successful. The file will have been created and attached to
3005 * @file by the filesystem calling finish_open().
3007 * Returns 1 if the file was looked up only or didn't need creating. The
3008 * caller will need to perform the open themselves. @path will have been
3009 * updated to point to the new dentry. This may be negative.
3011 * Returns an error code otherwise.
3013 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3014 struct path *path, struct file *file,
3015 const struct open_flags *op,
3016 int open_flag, umode_t mode,
3019 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3020 struct inode *dir = nd->path.dentry->d_inode;
3023 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3024 open_flag &= ~O_TRUNC;
3026 if (nd->flags & LOOKUP_DIRECTORY)
3027 open_flag |= O_DIRECTORY;
3029 file->f_path.dentry = DENTRY_NOT_SET;
3030 file->f_path.mnt = nd->path.mnt;
3031 error = dir->i_op->atomic_open(dir, dentry, file,
3032 open_to_namei_flags(open_flag),
3034 d_lookup_done(dentry);
3037 * We didn't have the inode before the open, so check open
3040 int acc_mode = op->acc_mode;
3041 if (*opened & FILE_CREATED) {
3042 WARN_ON(!(open_flag & O_CREAT));
3043 fsnotify_create(dir, dentry);
3046 error = may_open(&file->f_path, acc_mode, open_flag);
3047 if (WARN_ON(error > 0))
3049 } else if (error > 0) {
3050 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3053 if (file->f_path.dentry) {
3055 dentry = file->f_path.dentry;
3057 if (*opened & FILE_CREATED)
3058 fsnotify_create(dir, dentry);
3059 if (unlikely(d_is_negative(dentry))) {
3062 path->dentry = dentry;
3063 path->mnt = nd->path.mnt;
3073 * Look up and maybe create and open the last component.
3075 * Must be called with i_mutex held on parent.
3077 * Returns 0 if the file was successfully atomically created (if necessary) and
3078 * opened. In this case the file will be returned attached to @file.
3080 * Returns 1 if the file was not completely opened at this time, though lookups
3081 * and creations will have been performed and the dentry returned in @path will
3082 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3083 * specified then a negative dentry may be returned.
3085 * An error code is returned otherwise.
3087 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3088 * cleared otherwise prior to returning.
3090 static int lookup_open(struct nameidata *nd, struct path *path,
3092 const struct open_flags *op,
3093 bool got_write, int *opened)
3095 struct dentry *dir = nd->path.dentry;
3096 struct inode *dir_inode = dir->d_inode;
3097 int open_flag = op->open_flag;
3098 struct dentry *dentry;
3099 int error, create_error = 0;
3100 umode_t mode = op->mode;
3101 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3103 if (unlikely(IS_DEADDIR(dir_inode)))
3106 *opened &= ~FILE_CREATED;
3107 dentry = d_lookup(dir, &nd->last);
3110 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3112 return PTR_ERR(dentry);
3114 if (d_in_lookup(dentry))
3117 error = d_revalidate(dentry, nd->flags);
3118 if (likely(error > 0))
3122 d_invalidate(dentry);
3126 if (dentry->d_inode) {
3127 /* Cached positive dentry: will open in f_op->open */
3132 * Checking write permission is tricky, bacuse we don't know if we are
3133 * going to actually need it: O_CREAT opens should work as long as the
3134 * file exists. But checking existence breaks atomicity. The trick is
3135 * to check access and if not granted clear O_CREAT from the flags.
3137 * Another problem is returing the "right" error value (e.g. for an
3138 * O_EXCL open we want to return EEXIST not EROFS).
3140 if (open_flag & O_CREAT) {
3141 if (!IS_POSIXACL(dir->d_inode))
3142 mode &= ~current_umask();
3143 if (unlikely(!got_write)) {
3144 create_error = -EROFS;
3145 open_flag &= ~O_CREAT;
3146 if (open_flag & (O_EXCL | O_TRUNC))
3148 /* No side effects, safe to clear O_CREAT */
3150 create_error = may_o_create(&nd->path, dentry, mode);
3152 open_flag &= ~O_CREAT;
3153 if (open_flag & O_EXCL)
3157 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3158 unlikely(!got_write)) {
3160 * No O_CREATE -> atomicity not a requirement -> fall
3161 * back to lookup + open
3166 if (dir_inode->i_op->atomic_open) {
3167 error = atomic_open(nd, dentry, path, file, op, open_flag,
3169 if (unlikely(error == -ENOENT) && create_error)
3170 error = create_error;
3175 if (d_in_lookup(dentry)) {
3176 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3178 d_lookup_done(dentry);
3179 if (unlikely(res)) {
3181 error = PTR_ERR(res);
3189 /* Negative dentry, just create the file */
3190 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3191 *opened |= FILE_CREATED;
3192 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3193 if (!dir_inode->i_op->create) {
3197 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3198 open_flag & O_EXCL);
3201 fsnotify_create(dir_inode, dentry);
3203 if (unlikely(create_error) && !dentry->d_inode) {
3204 error = create_error;
3208 path->dentry = dentry;
3209 path->mnt = nd->path.mnt;
3218 * Handle the last step of open()
3220 static int do_last(struct nameidata *nd,
3221 struct file *file, const struct open_flags *op,
3224 struct dentry *dir = nd->path.dentry;
3225 int open_flag = op->open_flag;
3226 bool will_truncate = (open_flag & O_TRUNC) != 0;
3227 bool got_write = false;
3228 int acc_mode = op->acc_mode;
3230 struct inode *inode;
3234 nd->flags &= ~LOOKUP_PARENT;
3235 nd->flags |= op->intent;
3237 if (nd->last_type != LAST_NORM) {
3238 error = handle_dots(nd, nd->last_type);
3239 if (unlikely(error))
3244 if (!(open_flag & O_CREAT)) {
3245 if (nd->last.name[nd->last.len])
3246 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3247 /* we _can_ be in RCU mode here */
3248 error = lookup_fast(nd, &path, &inode, &seq);
3249 if (likely(error > 0))
3255 BUG_ON(nd->inode != dir->d_inode);
3256 BUG_ON(nd->flags & LOOKUP_RCU);
3258 /* create side of things */
3260 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3261 * has been cleared when we got to the last component we are
3264 error = complete_walk(nd);
3268 audit_inode(nd->name, dir, LOOKUP_PARENT);
3269 /* trailing slashes? */
3270 if (unlikely(nd->last.name[nd->last.len]))
3274 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3275 error = mnt_want_write(nd->path.mnt);
3279 * do _not_ fail yet - we might not need that or fail with
3280 * a different error; let lookup_open() decide; we'll be
3281 * dropping this one anyway.
3284 if (open_flag & O_CREAT)
3285 inode_lock(dir->d_inode);
3287 inode_lock_shared(dir->d_inode);
3288 error = lookup_open(nd, &path, file, op, got_write, opened);
3289 if (open_flag & O_CREAT)
3290 inode_unlock(dir->d_inode);
3292 inode_unlock_shared(dir->d_inode);
3298 if ((*opened & FILE_CREATED) ||
3299 !S_ISREG(file_inode(file)->i_mode))
3300 will_truncate = false;
3302 audit_inode(nd->name, file->f_path.dentry, 0);
3306 if (*opened & FILE_CREATED) {
3307 /* Don't check for write permission, don't truncate */
3308 open_flag &= ~O_TRUNC;
3309 will_truncate = false;
3311 path_to_nameidata(&path, nd);
3312 goto finish_open_created;
3316 * If atomic_open() acquired write access it is dropped now due to
3317 * possible mount and symlink following (this might be optimized away if
3321 mnt_drop_write(nd->path.mnt);
3325 error = follow_managed(&path, nd);
3326 if (unlikely(error < 0))
3329 if (unlikely(d_is_negative(path.dentry))) {
3330 path_to_nameidata(&path, nd);
3335 * create/update audit record if it already exists.
3337 audit_inode(nd->name, path.dentry, 0);
3339 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3340 path_to_nameidata(&path, nd);
3344 seq = 0; /* out of RCU mode, so the value doesn't matter */
3345 inode = d_backing_inode(path.dentry);
3347 error = step_into(nd, &path, 0, inode, seq);
3348 if (unlikely(error))
3351 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3352 error = complete_walk(nd);
3355 audit_inode(nd->name, nd->path.dentry, 0);
3357 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3360 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3362 if (!d_is_reg(nd->path.dentry))
3363 will_truncate = false;
3365 if (will_truncate) {
3366 error = mnt_want_write(nd->path.mnt);
3371 finish_open_created:
3372 error = may_open(&nd->path, acc_mode, open_flag);
3375 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3376 error = vfs_open(&nd->path, file, current_cred());
3379 *opened |= FILE_OPENED;
3381 error = open_check_o_direct(file);
3383 error = ima_file_check(file, op->acc_mode, *opened);
3384 if (!error && will_truncate)
3385 error = handle_truncate(file);
3387 if (unlikely(error) && (*opened & FILE_OPENED))
3389 if (unlikely(error > 0)) {
3394 mnt_drop_write(nd->path.mnt);
3398 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3400 struct dentry *child = NULL;
3401 struct inode *dir = dentry->d_inode;
3402 struct inode *inode;
3405 /* we want directory to be writable */
3406 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3409 error = -EOPNOTSUPP;
3410 if (!dir->i_op->tmpfile)
3413 child = d_alloc(dentry, &slash_name);
3414 if (unlikely(!child))
3416 error = dir->i_op->tmpfile(dir, child, mode);
3420 inode = child->d_inode;
3421 if (unlikely(!inode))
3423 if (!(open_flag & O_EXCL)) {
3424 spin_lock(&inode->i_lock);
3425 inode->i_state |= I_LINKABLE;
3426 spin_unlock(&inode->i_lock);
3432 return ERR_PTR(error);
3434 EXPORT_SYMBOL(vfs_tmpfile);
3436 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3437 const struct open_flags *op,
3438 struct file *file, int *opened)
3440 struct dentry *child;
3442 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3443 if (unlikely(error))
3445 error = mnt_want_write(path.mnt);
3446 if (unlikely(error))
3448 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3449 error = PTR_ERR(child);
3453 path.dentry = child;
3454 audit_inode(nd->name, child, 0);
3455 /* Don't check for other permissions, the inode was just created */
3456 error = may_open(&path, 0, op->open_flag);
3459 file->f_path.mnt = path.mnt;
3460 error = finish_open(file, child, NULL, opened);
3463 error = open_check_o_direct(file);
3467 mnt_drop_write(path.mnt);
3473 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3476 int error = path_lookupat(nd, flags, &path);
3478 audit_inode(nd->name, path.dentry, 0);
3479 error = vfs_open(&path, file, current_cred());
3485 static struct file *path_openat(struct nameidata *nd,
3486 const struct open_flags *op, unsigned flags)
3493 file = get_empty_filp();
3497 file->f_flags = op->open_flag;
3499 if (unlikely(file->f_flags & __O_TMPFILE)) {
3500 error = do_tmpfile(nd, flags, op, file, &opened);
3504 if (unlikely(file->f_flags & O_PATH)) {
3505 error = do_o_path(nd, flags, file);
3507 opened |= FILE_OPENED;
3511 s = path_init(nd, flags);
3516 while (!(error = link_path_walk(s, nd)) &&
3517 (error = do_last(nd, file, op, &opened)) > 0) {
3518 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3519 s = trailing_symlink(nd);
3527 if (!(opened & FILE_OPENED)) {
3531 if (unlikely(error)) {
3532 if (error == -EOPENSTALE) {
3533 if (flags & LOOKUP_RCU)
3538 file = ERR_PTR(error);
3543 struct file *do_filp_open(int dfd, struct filename *pathname,
3544 const struct open_flags *op)
3546 struct nameidata nd;
3547 int flags = op->lookup_flags;
3550 set_nameidata(&nd, dfd, pathname);
3551 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3552 if (unlikely(filp == ERR_PTR(-ECHILD)))
3553 filp = path_openat(&nd, op, flags);
3554 if (unlikely(filp == ERR_PTR(-ESTALE)))
3555 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3556 restore_nameidata();
3560 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3561 const char *name, const struct open_flags *op)
3563 struct nameidata nd;
3565 struct filename *filename;
3566 int flags = op->lookup_flags | LOOKUP_ROOT;
3569 nd.root.dentry = dentry;
3571 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3572 return ERR_PTR(-ELOOP);
3574 filename = getname_kernel(name);
3575 if (IS_ERR(filename))
3576 return ERR_CAST(filename);
3578 set_nameidata(&nd, -1, filename);
3579 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3580 if (unlikely(file == ERR_PTR(-ECHILD)))
3581 file = path_openat(&nd, op, flags);
3582 if (unlikely(file == ERR_PTR(-ESTALE)))
3583 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3584 restore_nameidata();
3589 static struct dentry *filename_create(int dfd, struct filename *name,
3590 struct path *path, unsigned int lookup_flags)
3592 struct dentry *dentry = ERR_PTR(-EEXIST);
3597 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3600 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3601 * other flags passed in are ignored!
3603 lookup_flags &= LOOKUP_REVAL;
3605 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3607 return ERR_CAST(name);
3610 * Yucky last component or no last component at all?
3611 * (foo/., foo/.., /////)
3613 if (unlikely(type != LAST_NORM))
3616 /* don't fail immediately if it's r/o, at least try to report other errors */
3617 err2 = mnt_want_write(path->mnt);
3619 * Do the final lookup.
3621 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3622 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3623 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3628 if (d_is_positive(dentry))
3632 * Special case - lookup gave negative, but... we had foo/bar/
3633 * From the vfs_mknod() POV we just have a negative dentry -
3634 * all is fine. Let's be bastards - you had / on the end, you've
3635 * been asking for (non-existent) directory. -ENOENT for you.
3637 if (unlikely(!is_dir && last.name[last.len])) {
3641 if (unlikely(err2)) {
3649 dentry = ERR_PTR(error);
3651 inode_unlock(path->dentry->d_inode);
3653 mnt_drop_write(path->mnt);
3660 struct dentry *kern_path_create(int dfd, const char *pathname,
3661 struct path *path, unsigned int lookup_flags)
3663 return filename_create(dfd, getname_kernel(pathname),
3664 path, lookup_flags);
3666 EXPORT_SYMBOL(kern_path_create);
3668 void done_path_create(struct path *path, struct dentry *dentry)
3671 inode_unlock(path->dentry->d_inode);
3672 mnt_drop_write(path->mnt);
3675 EXPORT_SYMBOL(done_path_create);
3677 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3678 struct path *path, unsigned int lookup_flags)
3680 return filename_create(dfd, getname(pathname), path, lookup_flags);
3682 EXPORT_SYMBOL(user_path_create);
3684 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3686 int error = may_create(dir, dentry);
3691 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3694 if (!dir->i_op->mknod)
3697 error = devcgroup_inode_mknod(mode, dev);
3701 error = security_inode_mknod(dir, dentry, mode, dev);
3705 error = dir->i_op->mknod(dir, dentry, mode, dev);
3707 fsnotify_create(dir, dentry);
3710 EXPORT_SYMBOL(vfs_mknod);
3712 static int may_mknod(umode_t mode)
3714 switch (mode & S_IFMT) {
3720 case 0: /* zero mode translates to S_IFREG */
3729 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3732 struct dentry *dentry;
3735 unsigned int lookup_flags = 0;
3737 error = may_mknod(mode);
3741 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3743 return PTR_ERR(dentry);
3745 if (!IS_POSIXACL(path.dentry->d_inode))
3746 mode &= ~current_umask();
3747 error = security_path_mknod(&path, dentry, mode, dev);
3750 switch (mode & S_IFMT) {
3751 case 0: case S_IFREG:
3752 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3754 ima_post_path_mknod(dentry);
3756 case S_IFCHR: case S_IFBLK:
3757 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3758 new_decode_dev(dev));
3760 case S_IFIFO: case S_IFSOCK:
3761 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3765 done_path_create(&path, dentry);
3766 if (retry_estale(error, lookup_flags)) {
3767 lookup_flags |= LOOKUP_REVAL;
3773 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3775 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3778 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3780 int error = may_create(dir, dentry);
3781 unsigned max_links = dir->i_sb->s_max_links;
3786 if (!dir->i_op->mkdir)
3789 mode &= (S_IRWXUGO|S_ISVTX);
3790 error = security_inode_mkdir(dir, dentry, mode);
3794 if (max_links && dir->i_nlink >= max_links)
3797 error = dir->i_op->mkdir(dir, dentry, mode);
3799 fsnotify_mkdir(dir, dentry);
3802 EXPORT_SYMBOL(vfs_mkdir);
3804 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3806 struct dentry *dentry;
3809 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3812 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3814 return PTR_ERR(dentry);
3816 if (!IS_POSIXACL(path.dentry->d_inode))
3817 mode &= ~current_umask();
3818 error = security_path_mkdir(&path, dentry, mode);
3820 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3821 done_path_create(&path, dentry);
3822 if (retry_estale(error, lookup_flags)) {
3823 lookup_flags |= LOOKUP_REVAL;
3829 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3831 return sys_mkdirat(AT_FDCWD, pathname, mode);
3834 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3836 int error = may_delete(dir, dentry, 1);
3841 if (!dir->i_op->rmdir)
3845 inode_lock(dentry->d_inode);
3848 if (is_local_mountpoint(dentry))
3851 error = security_inode_rmdir(dir, dentry);
3855 shrink_dcache_parent(dentry);
3856 error = dir->i_op->rmdir(dir, dentry);
3860 dentry->d_inode->i_flags |= S_DEAD;
3862 detach_mounts(dentry);
3865 inode_unlock(dentry->d_inode);
3871 EXPORT_SYMBOL(vfs_rmdir);
3873 static long do_rmdir(int dfd, const char __user *pathname)
3876 struct filename *name;
3877 struct dentry *dentry;
3881 unsigned int lookup_flags = 0;
3883 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3884 &path, &last, &type);
3886 return PTR_ERR(name);
3900 error = mnt_want_write(path.mnt);
3904 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3905 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3906 error = PTR_ERR(dentry);
3909 if (!dentry->d_inode) {
3913 error = security_path_rmdir(&path, dentry);
3916 error = vfs_rmdir(path.dentry->d_inode, dentry);
3920 inode_unlock(path.dentry->d_inode);
3921 mnt_drop_write(path.mnt);
3925 if (retry_estale(error, lookup_flags)) {
3926 lookup_flags |= LOOKUP_REVAL;
3932 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3934 return do_rmdir(AT_FDCWD, pathname);
3938 * vfs_unlink - unlink a filesystem object
3939 * @dir: parent directory
3941 * @delegated_inode: returns victim inode, if the inode is delegated.
3943 * The caller must hold dir->i_mutex.
3945 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3946 * return a reference to the inode in delegated_inode. The caller
3947 * should then break the delegation on that inode and retry. Because
3948 * breaking a delegation may take a long time, the caller should drop
3949 * dir->i_mutex before doing so.
3951 * Alternatively, a caller may pass NULL for delegated_inode. This may
3952 * be appropriate for callers that expect the underlying filesystem not
3953 * to be NFS exported.
3955 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3957 struct inode *target = dentry->d_inode;
3958 int error = may_delete(dir, dentry, 0);
3963 if (!dir->i_op->unlink)
3967 if (is_local_mountpoint(dentry))
3970 error = security_inode_unlink(dir, dentry);
3972 error = try_break_deleg(target, delegated_inode);
3975 error = dir->i_op->unlink(dir, dentry);
3978 detach_mounts(dentry);
3983 inode_unlock(target);
3985 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3986 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3987 fsnotify_link_count(target);
3993 EXPORT_SYMBOL(vfs_unlink);
3996 * Make sure that the actual truncation of the file will occur outside its
3997 * directory's i_mutex. Truncate can take a long time if there is a lot of
3998 * writeout happening, and we don't want to prevent access to the directory
3999 * while waiting on the I/O.
4001 long do_unlinkat(int dfd, struct filename *name)
4004 struct dentry *dentry;
4008 struct inode *inode = NULL;
4009 struct inode *delegated_inode = NULL;
4010 unsigned int lookup_flags = 0;
4012 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4014 return PTR_ERR(name);
4017 if (type != LAST_NORM)
4020 error = mnt_want_write(path.mnt);
4024 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4025 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4026 error = PTR_ERR(dentry);
4027 if (!IS_ERR(dentry)) {
4028 /* Why not before? Because we want correct error value */
4029 if (last.name[last.len])
4031 inode = dentry->d_inode;
4032 if (d_is_negative(dentry))
4035 error = security_path_unlink(&path, dentry);
4038 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4042 inode_unlock(path.dentry->d_inode);
4044 iput(inode); /* truncate the inode here */
4046 if (delegated_inode) {
4047 error = break_deleg_wait(&delegated_inode);
4051 mnt_drop_write(path.mnt);
4054 if (retry_estale(error, lookup_flags)) {
4055 lookup_flags |= LOOKUP_REVAL;
4063 if (d_is_negative(dentry))
4065 else if (d_is_dir(dentry))
4072 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4074 if ((flag & ~AT_REMOVEDIR) != 0)
4077 if (flag & AT_REMOVEDIR)
4078 return do_rmdir(dfd, pathname);
4080 return do_unlinkat(dfd, getname(pathname));
4083 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4085 return do_unlinkat(AT_FDCWD, getname(pathname));
4088 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4090 int error = may_create(dir, dentry);
4095 if (!dir->i_op->symlink)
4098 error = security_inode_symlink(dir, dentry, oldname);
4102 error = dir->i_op->symlink(dir, dentry, oldname);
4104 fsnotify_create(dir, dentry);
4107 EXPORT_SYMBOL(vfs_symlink);
4109 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4110 int, newdfd, const char __user *, newname)
4113 struct filename *from;
4114 struct dentry *dentry;
4116 unsigned int lookup_flags = 0;
4118 from = getname(oldname);
4120 return PTR_ERR(from);
4122 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4123 error = PTR_ERR(dentry);
4127 error = security_path_symlink(&path, dentry, from->name);
4129 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4130 done_path_create(&path, dentry);
4131 if (retry_estale(error, lookup_flags)) {
4132 lookup_flags |= LOOKUP_REVAL;
4140 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4142 return sys_symlinkat(oldname, AT_FDCWD, newname);
4146 * vfs_link - create a new link
4147 * @old_dentry: object to be linked
4149 * @new_dentry: where to create the new link
4150 * @delegated_inode: returns inode needing a delegation break
4152 * The caller must hold dir->i_mutex
4154 * If vfs_link discovers a delegation on the to-be-linked file in need
4155 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4156 * inode in delegated_inode. The caller should then break the delegation
4157 * and retry. Because breaking a delegation may take a long time, the
4158 * caller should drop the i_mutex before doing so.
4160 * Alternatively, a caller may pass NULL for delegated_inode. This may
4161 * be appropriate for callers that expect the underlying filesystem not
4162 * to be NFS exported.
4164 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4166 struct inode *inode = old_dentry->d_inode;
4167 unsigned max_links = dir->i_sb->s_max_links;
4173 error = may_create(dir, new_dentry);
4177 if (dir->i_sb != inode->i_sb)
4181 * A link to an append-only or immutable file cannot be created.
4183 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4186 * Updating the link count will likely cause i_uid and i_gid to
4187 * be writen back improperly if their true value is unknown to
4190 if (HAS_UNMAPPED_ID(inode))
4192 if (!dir->i_op->link)
4194 if (S_ISDIR(inode->i_mode))
4197 error = security_inode_link(old_dentry, dir, new_dentry);
4202 /* Make sure we don't allow creating hardlink to an unlinked file */
4203 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4205 else if (max_links && inode->i_nlink >= max_links)
4208 error = try_break_deleg(inode, delegated_inode);
4210 error = dir->i_op->link(old_dentry, dir, new_dentry);
4213 if (!error && (inode->i_state & I_LINKABLE)) {
4214 spin_lock(&inode->i_lock);
4215 inode->i_state &= ~I_LINKABLE;
4216 spin_unlock(&inode->i_lock);
4218 inode_unlock(inode);
4220 fsnotify_link(dir, inode, new_dentry);
4223 EXPORT_SYMBOL(vfs_link);
4226 * Hardlinks are often used in delicate situations. We avoid
4227 * security-related surprises by not following symlinks on the
4230 * We don't follow them on the oldname either to be compatible
4231 * with linux 2.0, and to avoid hard-linking to directories
4232 * and other special files. --ADM
4234 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4235 int, newdfd, const char __user *, newname, int, flags)
4237 struct dentry *new_dentry;
4238 struct path old_path, new_path;
4239 struct inode *delegated_inode = NULL;
4243 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4246 * To use null names we require CAP_DAC_READ_SEARCH
4247 * This ensures that not everyone will be able to create
4248 * handlink using the passed filedescriptor.
4250 if (flags & AT_EMPTY_PATH) {
4251 if (!capable(CAP_DAC_READ_SEARCH))
4256 if (flags & AT_SYMLINK_FOLLOW)
4257 how |= LOOKUP_FOLLOW;
4259 error = user_path_at(olddfd, oldname, how, &old_path);
4263 new_dentry = user_path_create(newdfd, newname, &new_path,
4264 (how & LOOKUP_REVAL));
4265 error = PTR_ERR(new_dentry);
4266 if (IS_ERR(new_dentry))
4270 if (old_path.mnt != new_path.mnt)
4272 error = may_linkat(&old_path);
4273 if (unlikely(error))
4275 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4278 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4280 done_path_create(&new_path, new_dentry);
4281 if (delegated_inode) {
4282 error = break_deleg_wait(&delegated_inode);
4284 path_put(&old_path);
4288 if (retry_estale(error, how)) {
4289 path_put(&old_path);
4290 how |= LOOKUP_REVAL;
4294 path_put(&old_path);
4299 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4301 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4305 * vfs_rename - rename a filesystem object
4306 * @old_dir: parent of source
4307 * @old_dentry: source
4308 * @new_dir: parent of destination
4309 * @new_dentry: destination
4310 * @delegated_inode: returns an inode needing a delegation break
4311 * @flags: rename flags
4313 * The caller must hold multiple mutexes--see lock_rename()).
4315 * If vfs_rename discovers a delegation in need of breaking at either
4316 * the source or destination, it will return -EWOULDBLOCK and return a
4317 * reference to the inode in delegated_inode. The caller should then
4318 * break the delegation and retry. Because breaking a delegation may
4319 * take a long time, the caller should drop all locks before doing
4322 * Alternatively, a caller may pass NULL for delegated_inode. This may
4323 * be appropriate for callers that expect the underlying filesystem not
4324 * to be NFS exported.
4326 * The worst of all namespace operations - renaming directory. "Perverted"
4327 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4330 * a) we can get into loop creation.
4331 * b) race potential - two innocent renames can create a loop together.
4332 * That's where 4.4 screws up. Current fix: serialization on
4333 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4335 * c) we have to lock _four_ objects - parents and victim (if it exists),
4336 * and source (if it is not a directory).
4337 * And that - after we got ->i_mutex on parents (until then we don't know
4338 * whether the target exists). Solution: try to be smart with locking
4339 * order for inodes. We rely on the fact that tree topology may change
4340 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4341 * move will be locked. Thus we can rank directories by the tree
4342 * (ancestors first) and rank all non-directories after them.
4343 * That works since everybody except rename does "lock parent, lookup,
4344 * lock child" and rename is under ->s_vfs_rename_mutex.
4345 * HOWEVER, it relies on the assumption that any object with ->lookup()
4346 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4347 * we'd better make sure that there's no link(2) for them.
4348 * d) conversion from fhandle to dentry may come in the wrong moment - when
4349 * we are removing the target. Solution: we will have to grab ->i_mutex
4350 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4351 * ->i_mutex on parents, which works but leads to some truly excessive
4354 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4355 struct inode *new_dir, struct dentry *new_dentry,
4356 struct inode **delegated_inode, unsigned int flags)
4359 bool is_dir = d_is_dir(old_dentry);
4360 struct inode *source = old_dentry->d_inode;
4361 struct inode *target = new_dentry->d_inode;
4362 bool new_is_dir = false;
4363 unsigned max_links = new_dir->i_sb->s_max_links;
4364 struct name_snapshot old_name;
4366 if (source == target)
4369 error = may_delete(old_dir, old_dentry, is_dir);
4374 error = may_create(new_dir, new_dentry);
4376 new_is_dir = d_is_dir(new_dentry);
4378 if (!(flags & RENAME_EXCHANGE))
4379 error = may_delete(new_dir, new_dentry, is_dir);
4381 error = may_delete(new_dir, new_dentry, new_is_dir);
4386 if (!old_dir->i_op->rename)
4390 * If we are going to change the parent - check write permissions,
4391 * we'll need to flip '..'.
4393 if (new_dir != old_dir) {
4395 error = inode_permission(source, MAY_WRITE);
4399 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4400 error = inode_permission(target, MAY_WRITE);
4406 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4411 take_dentry_name_snapshot(&old_name, old_dentry);
4413 if (!is_dir || (flags & RENAME_EXCHANGE))
4414 lock_two_nondirectories(source, target);
4419 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4422 if (max_links && new_dir != old_dir) {
4424 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4426 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4427 old_dir->i_nlink >= max_links)
4430 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4431 shrink_dcache_parent(new_dentry);
4433 error = try_break_deleg(source, delegated_inode);
4437 if (target && !new_is_dir) {
4438 error = try_break_deleg(target, delegated_inode);
4442 error = old_dir->i_op->rename(old_dir, old_dentry,
4443 new_dir, new_dentry, flags);
4447 if (!(flags & RENAME_EXCHANGE) && target) {
4449 target->i_flags |= S_DEAD;
4450 dont_mount(new_dentry);
4451 detach_mounts(new_dentry);
4453 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4454 if (!(flags & RENAME_EXCHANGE))
4455 d_move(old_dentry, new_dentry);
4457 d_exchange(old_dentry, new_dentry);
4460 if (!is_dir || (flags & RENAME_EXCHANGE))
4461 unlock_two_nondirectories(source, target);
4463 inode_unlock(target);
4466 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4467 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4468 if (flags & RENAME_EXCHANGE) {
4469 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4470 new_is_dir, NULL, new_dentry);
4473 release_dentry_name_snapshot(&old_name);
4477 EXPORT_SYMBOL(vfs_rename);
4479 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4480 int, newdfd, const char __user *, newname, unsigned int, flags)
4482 struct dentry *old_dentry, *new_dentry;
4483 struct dentry *trap;
4484 struct path old_path, new_path;
4485 struct qstr old_last, new_last;
4486 int old_type, new_type;
4487 struct inode *delegated_inode = NULL;
4488 struct filename *from;
4489 struct filename *to;
4490 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4491 bool should_retry = false;
4494 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4497 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4498 (flags & RENAME_EXCHANGE))
4501 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4504 if (flags & RENAME_EXCHANGE)
4508 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4509 &old_path, &old_last, &old_type);
4511 error = PTR_ERR(from);
4515 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4516 &new_path, &new_last, &new_type);
4518 error = PTR_ERR(to);
4523 if (old_path.mnt != new_path.mnt)
4527 if (old_type != LAST_NORM)
4530 if (flags & RENAME_NOREPLACE)
4532 if (new_type != LAST_NORM)
4535 error = mnt_want_write(old_path.mnt);
4540 trap = lock_rename(new_path.dentry, old_path.dentry);
4542 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4543 error = PTR_ERR(old_dentry);
4544 if (IS_ERR(old_dentry))
4546 /* source must exist */
4548 if (d_is_negative(old_dentry))
4550 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4551 error = PTR_ERR(new_dentry);
4552 if (IS_ERR(new_dentry))
4555 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4557 if (flags & RENAME_EXCHANGE) {
4559 if (d_is_negative(new_dentry))
4562 if (!d_is_dir(new_dentry)) {
4564 if (new_last.name[new_last.len])
4568 /* unless the source is a directory trailing slashes give -ENOTDIR */
4569 if (!d_is_dir(old_dentry)) {
4571 if (old_last.name[old_last.len])
4573 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4576 /* source should not be ancestor of target */
4578 if (old_dentry == trap)
4580 /* target should not be an ancestor of source */
4581 if (!(flags & RENAME_EXCHANGE))
4583 if (new_dentry == trap)
4586 error = security_path_rename(&old_path, old_dentry,
4587 &new_path, new_dentry, flags);
4590 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4591 new_path.dentry->d_inode, new_dentry,
4592 &delegated_inode, flags);
4598 unlock_rename(new_path.dentry, old_path.dentry);
4599 if (delegated_inode) {
4600 error = break_deleg_wait(&delegated_inode);
4604 mnt_drop_write(old_path.mnt);
4606 if (retry_estale(error, lookup_flags))
4607 should_retry = true;
4608 path_put(&new_path);
4611 path_put(&old_path);
4614 should_retry = false;
4615 lookup_flags |= LOOKUP_REVAL;
4622 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4623 int, newdfd, const char __user *, newname)
4625 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4628 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4630 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4633 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4635 int error = may_create(dir, dentry);
4639 if (!dir->i_op->mknod)
4642 return dir->i_op->mknod(dir, dentry,
4643 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4645 EXPORT_SYMBOL(vfs_whiteout);
4647 int readlink_copy(char __user *buffer, int buflen, const char *link)
4649 int len = PTR_ERR(link);
4654 if (len > (unsigned) buflen)
4656 if (copy_to_user(buffer, link, len))
4663 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4664 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4665 * for any given inode is up to filesystem.
4667 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4670 DEFINE_DELAYED_CALL(done);
4671 struct inode *inode = d_inode(dentry);
4672 const char *link = 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);
4686 * vfs_readlink - copy symlink body into userspace buffer
4687 * @dentry: dentry on which to get symbolic link
4688 * @buffer: user memory pointer
4689 * @buflen: size of buffer
4691 * Does not touch atime. That's up to the caller if necessary
4693 * Does not call security hook.
4695 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4697 struct inode *inode = d_inode(dentry);
4699 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4700 if (unlikely(inode->i_op->readlink))
4701 return inode->i_op->readlink(dentry, buffer, buflen);
4703 if (!d_is_symlink(dentry))
4706 spin_lock(&inode->i_lock);
4707 inode->i_opflags |= IOP_DEFAULT_READLINK;
4708 spin_unlock(&inode->i_lock);
4711 return generic_readlink(dentry, buffer, buflen);
4713 EXPORT_SYMBOL(vfs_readlink);
4716 * vfs_get_link - get symlink body
4717 * @dentry: dentry on which to get symbolic link
4718 * @done: caller needs to free returned data with this
4720 * Calls security hook and i_op->get_link() on the supplied inode.
4722 * It does not touch atime. That's up to the caller if necessary.
4724 * Does not work on "special" symlinks like /proc/$$/fd/N
4726 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4728 const char *res = ERR_PTR(-EINVAL);
4729 struct inode *inode = d_inode(dentry);
4731 if (d_is_symlink(dentry)) {
4732 res = ERR_PTR(security_inode_readlink(dentry));
4734 res = inode->i_op->get_link(dentry, inode, done);
4738 EXPORT_SYMBOL(vfs_get_link);
4740 /* get the link contents into pagecache */
4741 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4742 struct delayed_call *callback)
4746 struct address_space *mapping = inode->i_mapping;
4749 page = find_get_page(mapping, 0);
4751 return ERR_PTR(-ECHILD);
4752 if (!PageUptodate(page)) {
4754 return ERR_PTR(-ECHILD);
4757 page = read_mapping_page(mapping, 0, NULL);
4761 set_delayed_call(callback, page_put_link, page);
4762 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4763 kaddr = page_address(page);
4764 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4768 EXPORT_SYMBOL(page_get_link);
4770 void page_put_link(void *arg)
4774 EXPORT_SYMBOL(page_put_link);
4776 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4778 DEFINE_DELAYED_CALL(done);
4779 int res = readlink_copy(buffer, buflen,
4780 page_get_link(dentry, d_inode(dentry),
4782 do_delayed_call(&done);
4785 EXPORT_SYMBOL(page_readlink);
4788 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4790 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4792 struct address_space *mapping = inode->i_mapping;
4796 unsigned int flags = 0;
4798 flags |= AOP_FLAG_NOFS;
4801 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4802 flags, &page, &fsdata);
4806 memcpy(page_address(page), symname, len-1);
4808 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4815 mark_inode_dirty(inode);
4820 EXPORT_SYMBOL(__page_symlink);
4822 int page_symlink(struct inode *inode, const char *symname, int len)
4824 return __page_symlink(inode, symname, len,
4825 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4827 EXPORT_SYMBOL(page_symlink);
4829 const struct inode_operations page_symlink_inode_operations = {
4830 .get_link = page_get_link,
4832 EXPORT_SYMBOL(page_symlink_inode_operations);