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 * sb_permission - Check superblock-level permissions
395 * @sb: Superblock of inode to check permission on
396 * @inode: Inode to check permission on
397 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
399 * Separate out file-system wide checks from inode-specific permission checks.
401 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
403 if (unlikely(mask & MAY_WRITE)) {
404 umode_t mode = inode->i_mode;
406 /* Nobody gets write access to a read-only fs. */
407 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
414 * inode_permission - Check for access rights to a given inode
415 * @inode: Inode to check permission on
416 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
418 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
419 * this, letting us set arbitrary permissions for filesystem access without
420 * changing the "normal" UIDs which are used for other things.
422 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
424 int inode_permission(struct inode *inode, int mask)
428 retval = sb_permission(inode->i_sb, inode, mask);
432 if (unlikely(mask & MAY_WRITE)) {
434 * Nobody gets write access to an immutable file.
436 if (IS_IMMUTABLE(inode))
440 * Updating mtime will likely cause i_uid and i_gid to be
441 * written back improperly if their true value is unknown
444 if (HAS_UNMAPPED_ID(inode))
448 retval = do_inode_permission(inode, mask);
452 retval = devcgroup_inode_permission(inode, mask);
456 return security_inode_permission(inode, mask);
458 EXPORT_SYMBOL(inode_permission);
461 * path_get - get a reference to a path
462 * @path: path to get the reference to
464 * Given a path increment the reference count to the dentry and the vfsmount.
466 void path_get(const struct path *path)
471 EXPORT_SYMBOL(path_get);
474 * path_put - put a reference to a path
475 * @path: path to put the reference to
477 * Given a path decrement the reference count to the dentry and the vfsmount.
479 void path_put(const struct path *path)
484 EXPORT_SYMBOL(path_put);
486 #define EMBEDDED_LEVELS 2
491 struct inode *inode; /* path.dentry.d_inode */
496 int total_link_count;
499 struct delayed_call done;
502 } *stack, internal[EMBEDDED_LEVELS];
503 struct filename *name;
504 struct nameidata *saved;
505 struct inode *link_inode;
508 } __randomize_layout;
510 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
512 struct nameidata *old = current->nameidata;
513 p->stack = p->internal;
516 p->total_link_count = old ? old->total_link_count : 0;
518 current->nameidata = p;
521 static void restore_nameidata(void)
523 struct nameidata *now = current->nameidata, *old = now->saved;
525 current->nameidata = old;
527 old->total_link_count = now->total_link_count;
528 if (now->stack != now->internal)
532 static int __nd_alloc_stack(struct nameidata *nd)
536 if (nd->flags & LOOKUP_RCU) {
537 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
542 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
547 memcpy(p, nd->internal, sizeof(nd->internal));
553 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
554 * @path: nameidate to verify
556 * Rename can sometimes move a file or directory outside of a bind
557 * mount, path_connected allows those cases to be detected.
559 static bool path_connected(const struct path *path)
561 struct vfsmount *mnt = path->mnt;
562 struct super_block *sb = mnt->mnt_sb;
564 /* Bind mounts and multi-root filesystems can have disconnected paths */
565 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
568 return is_subdir(path->dentry, mnt->mnt_root);
571 static inline int nd_alloc_stack(struct nameidata *nd)
573 if (likely(nd->depth != EMBEDDED_LEVELS))
575 if (likely(nd->stack != nd->internal))
577 return __nd_alloc_stack(nd);
580 static void drop_links(struct nameidata *nd)
584 struct saved *last = nd->stack + i;
585 do_delayed_call(&last->done);
586 clear_delayed_call(&last->done);
590 static void terminate_walk(struct nameidata *nd)
593 if (!(nd->flags & LOOKUP_RCU)) {
596 for (i = 0; i < nd->depth; i++)
597 path_put(&nd->stack[i].link);
598 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
603 nd->flags &= ~LOOKUP_RCU;
604 if (!(nd->flags & LOOKUP_ROOT))
611 /* path_put is needed afterwards regardless of success or failure */
612 static bool legitimize_path(struct nameidata *nd,
613 struct path *path, unsigned seq)
615 int res = __legitimize_mnt(path->mnt, nd->m_seq);
622 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
626 return !read_seqcount_retry(&path->dentry->d_seq, seq);
629 static bool legitimize_links(struct nameidata *nd)
632 for (i = 0; i < nd->depth; i++) {
633 struct saved *last = nd->stack + i;
634 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
644 * Path walking has 2 modes, rcu-walk and ref-walk (see
645 * Documentation/filesystems/path-lookup.txt). In situations when we can't
646 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
647 * normal reference counts on dentries and vfsmounts to transition to ref-walk
648 * mode. Refcounts are grabbed at the last known good point before rcu-walk
649 * got stuck, so ref-walk may continue from there. If this is not successful
650 * (eg. a seqcount has changed), then failure is returned and it's up to caller
651 * to restart the path walk from the beginning in ref-walk mode.
655 * unlazy_walk - try to switch to ref-walk mode.
656 * @nd: nameidata pathwalk data
657 * Returns: 0 on success, -ECHILD on failure
659 * unlazy_walk attempts to legitimize the current nd->path and nd->root
661 * Must be called from rcu-walk context.
662 * Nothing should touch nameidata between unlazy_walk() failure and
665 static int unlazy_walk(struct nameidata *nd)
667 struct dentry *parent = nd->path.dentry;
669 BUG_ON(!(nd->flags & LOOKUP_RCU));
671 nd->flags &= ~LOOKUP_RCU;
672 if (unlikely(!legitimize_links(nd)))
674 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
676 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
677 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
681 BUG_ON(nd->inode != parent->d_inode);
686 nd->path.dentry = NULL;
688 if (!(nd->flags & LOOKUP_ROOT))
696 * unlazy_child - try to switch to ref-walk mode.
697 * @nd: nameidata pathwalk data
698 * @dentry: child of nd->path.dentry
699 * @seq: seq number to check dentry against
700 * Returns: 0 on success, -ECHILD on failure
702 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
703 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
704 * @nd. Must be called from rcu-walk context.
705 * Nothing should touch nameidata between unlazy_child() failure and
708 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
710 BUG_ON(!(nd->flags & LOOKUP_RCU));
712 nd->flags &= ~LOOKUP_RCU;
713 if (unlikely(!legitimize_links(nd)))
715 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
717 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
721 * We need to move both the parent and the dentry from the RCU domain
722 * to be properly refcounted. And the sequence number in the dentry
723 * validates *both* dentry counters, since we checked the sequence
724 * number of the parent after we got the child sequence number. So we
725 * know the parent must still be valid if the child sequence number is
727 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
729 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
735 * Sequence counts matched. Now make sure that the root is
736 * still valid and get it if required.
738 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
739 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
752 nd->path.dentry = NULL;
756 if (!(nd->flags & LOOKUP_ROOT))
761 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
763 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
764 return dentry->d_op->d_revalidate(dentry, flags);
770 * complete_walk - successful completion of path walk
771 * @nd: pointer nameidata
773 * If we had been in RCU mode, drop out of it and legitimize nd->path.
774 * Revalidate the final result, unless we'd already done that during
775 * the path walk or the filesystem doesn't ask for it. Return 0 on
776 * success, -error on failure. In case of failure caller does not
777 * need to drop nd->path.
779 static int complete_walk(struct nameidata *nd)
781 struct dentry *dentry = nd->path.dentry;
784 if (nd->flags & LOOKUP_RCU) {
785 if (!(nd->flags & LOOKUP_ROOT))
787 if (unlikely(unlazy_walk(nd)))
791 if (likely(!(nd->flags & LOOKUP_JUMPED)))
794 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
797 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
807 static void set_root(struct nameidata *nd)
809 struct fs_struct *fs = current->fs;
811 if (nd->flags & LOOKUP_RCU) {
815 seq = read_seqcount_begin(&fs->seq);
817 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
818 } while (read_seqcount_retry(&fs->seq, seq));
820 get_fs_root(fs, &nd->root);
824 static void path_put_conditional(struct path *path, struct nameidata *nd)
827 if (path->mnt != nd->path.mnt)
831 static inline void path_to_nameidata(const struct path *path,
832 struct nameidata *nd)
834 if (!(nd->flags & LOOKUP_RCU)) {
835 dput(nd->path.dentry);
836 if (nd->path.mnt != path->mnt)
837 mntput(nd->path.mnt);
839 nd->path.mnt = path->mnt;
840 nd->path.dentry = path->dentry;
843 static int nd_jump_root(struct nameidata *nd)
845 if (nd->flags & LOOKUP_RCU) {
849 nd->inode = d->d_inode;
850 nd->seq = nd->root_seq;
851 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
857 nd->inode = nd->path.dentry->d_inode;
859 nd->flags |= LOOKUP_JUMPED;
864 * Helper to directly jump to a known parsed path from ->get_link,
865 * caller must have taken a reference to path beforehand.
867 void nd_jump_link(struct path *path)
869 struct nameidata *nd = current->nameidata;
873 nd->inode = nd->path.dentry->d_inode;
874 nd->flags |= LOOKUP_JUMPED;
877 static inline void put_link(struct nameidata *nd)
879 struct saved *last = nd->stack + --nd->depth;
880 do_delayed_call(&last->done);
881 if (!(nd->flags & LOOKUP_RCU))
882 path_put(&last->link);
885 int sysctl_protected_symlinks __read_mostly = 0;
886 int sysctl_protected_hardlinks __read_mostly = 0;
889 * may_follow_link - Check symlink following for unsafe situations
890 * @nd: nameidata pathwalk data
892 * In the case of the sysctl_protected_symlinks sysctl being enabled,
893 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
894 * in a sticky world-writable directory. This is to protect privileged
895 * processes from failing races against path names that may change out
896 * from under them by way of other users creating malicious symlinks.
897 * It will permit symlinks to be followed only when outside a sticky
898 * world-writable directory, or when the uid of the symlink and follower
899 * match, or when the directory owner matches the symlink's owner.
901 * Returns 0 if following the symlink is allowed, -ve on error.
903 static inline int may_follow_link(struct nameidata *nd)
905 const struct inode *inode;
906 const struct inode *parent;
909 if (!sysctl_protected_symlinks)
912 /* Allowed if owner and follower match. */
913 inode = nd->link_inode;
914 if (uid_eq(current_cred()->fsuid, inode->i_uid))
917 /* Allowed if parent directory not sticky and world-writable. */
919 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
922 /* Allowed if parent directory and link owner match. */
923 puid = parent->i_uid;
924 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
927 if (nd->flags & LOOKUP_RCU)
930 audit_log_link_denied("follow_link", &nd->stack[0].link);
935 * safe_hardlink_source - Check for safe hardlink conditions
936 * @inode: the source inode to hardlink from
938 * Return false if at least one of the following conditions:
939 * - inode is not a regular file
941 * - inode is setgid and group-exec
942 * - access failure for read and write
944 * Otherwise returns true.
946 static bool safe_hardlink_source(struct inode *inode)
948 umode_t mode = inode->i_mode;
950 /* Special files should not get pinned to the filesystem. */
954 /* Setuid files should not get pinned to the filesystem. */
958 /* Executable setgid files should not get pinned to the filesystem. */
959 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
962 /* Hardlinking to unreadable or unwritable sources is dangerous. */
963 if (inode_permission(inode, MAY_READ | MAY_WRITE))
970 * may_linkat - Check permissions for creating a hardlink
971 * @link: the source to hardlink from
973 * Block hardlink when all of:
974 * - sysctl_protected_hardlinks enabled
975 * - fsuid does not match inode
976 * - hardlink source is unsafe (see safe_hardlink_source() above)
977 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
979 * Returns 0 if successful, -ve on error.
981 static int may_linkat(struct path *link)
985 if (!sysctl_protected_hardlinks)
988 inode = link->dentry->d_inode;
990 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
991 * otherwise, it must be a safe source.
993 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
996 audit_log_link_denied("linkat", link);
1000 static __always_inline
1001 const char *get_link(struct nameidata *nd)
1003 struct saved *last = nd->stack + nd->depth - 1;
1004 struct dentry *dentry = last->link.dentry;
1005 struct inode *inode = nd->link_inode;
1009 if (!(nd->flags & LOOKUP_RCU)) {
1010 touch_atime(&last->link);
1012 } else if (atime_needs_update_rcu(&last->link, inode)) {
1013 if (unlikely(unlazy_walk(nd)))
1014 return ERR_PTR(-ECHILD);
1015 touch_atime(&last->link);
1018 error = security_inode_follow_link(dentry, inode,
1019 nd->flags & LOOKUP_RCU);
1020 if (unlikely(error))
1021 return ERR_PTR(error);
1023 nd->last_type = LAST_BIND;
1024 res = inode->i_link;
1026 const char * (*get)(struct dentry *, struct inode *,
1027 struct delayed_call *);
1028 get = inode->i_op->get_link;
1029 if (nd->flags & LOOKUP_RCU) {
1030 res = get(NULL, inode, &last->done);
1031 if (res == ERR_PTR(-ECHILD)) {
1032 if (unlikely(unlazy_walk(nd)))
1033 return ERR_PTR(-ECHILD);
1034 res = get(dentry, inode, &last->done);
1037 res = get(dentry, inode, &last->done);
1039 if (IS_ERR_OR_NULL(res))
1045 if (unlikely(nd_jump_root(nd)))
1046 return ERR_PTR(-ECHILD);
1047 while (unlikely(*++res == '/'))
1056 * follow_up - Find the mountpoint of path's vfsmount
1058 * Given a path, find the mountpoint of its source file system.
1059 * Replace @path with the path of the mountpoint in the parent mount.
1062 * Return 1 if we went up a level and 0 if we were already at the
1065 int follow_up(struct path *path)
1067 struct mount *mnt = real_mount(path->mnt);
1068 struct mount *parent;
1069 struct dentry *mountpoint;
1071 read_seqlock_excl(&mount_lock);
1072 parent = mnt->mnt_parent;
1073 if (parent == mnt) {
1074 read_sequnlock_excl(&mount_lock);
1077 mntget(&parent->mnt);
1078 mountpoint = dget(mnt->mnt_mountpoint);
1079 read_sequnlock_excl(&mount_lock);
1081 path->dentry = mountpoint;
1083 path->mnt = &parent->mnt;
1086 EXPORT_SYMBOL(follow_up);
1089 * Perform an automount
1090 * - return -EISDIR to tell follow_managed() to stop and return the path we
1093 static int follow_automount(struct path *path, struct nameidata *nd,
1096 struct vfsmount *mnt;
1099 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1102 /* We don't want to mount if someone's just doing a stat -
1103 * unless they're stat'ing a directory and appended a '/' to
1106 * We do, however, want to mount if someone wants to open or
1107 * create a file of any type under the mountpoint, wants to
1108 * traverse through the mountpoint or wants to open the
1109 * mounted directory. Also, autofs may mark negative dentries
1110 * as being automount points. These will need the attentions
1111 * of the daemon to instantiate them before they can be used.
1113 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1114 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1115 path->dentry->d_inode)
1118 nd->total_link_count++;
1119 if (nd->total_link_count >= 40)
1122 mnt = path->dentry->d_op->d_automount(path);
1125 * The filesystem is allowed to return -EISDIR here to indicate
1126 * it doesn't want to automount. For instance, autofs would do
1127 * this so that its userspace daemon can mount on this dentry.
1129 * However, we can only permit this if it's a terminal point in
1130 * the path being looked up; if it wasn't then the remainder of
1131 * the path is inaccessible and we should say so.
1133 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1135 return PTR_ERR(mnt);
1138 if (!mnt) /* mount collision */
1141 if (!*need_mntput) {
1142 /* lock_mount() may release path->mnt on error */
1144 *need_mntput = true;
1146 err = finish_automount(mnt, path);
1150 /* Someone else made a mount here whilst we were busy */
1155 path->dentry = dget(mnt->mnt_root);
1164 * Handle a dentry that is managed in some way.
1165 * - Flagged for transit management (autofs)
1166 * - Flagged as mountpoint
1167 * - Flagged as automount point
1169 * This may only be called in refwalk mode.
1171 * Serialization is taken care of in namespace.c
1173 static int follow_managed(struct path *path, struct nameidata *nd)
1175 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1177 bool need_mntput = false;
1180 /* Given that we're not holding a lock here, we retain the value in a
1181 * local variable for each dentry as we look at it so that we don't see
1182 * the components of that value change under us */
1183 while (managed = READ_ONCE(path->dentry->d_flags),
1184 managed &= DCACHE_MANAGED_DENTRY,
1185 unlikely(managed != 0)) {
1186 /* Allow the filesystem to manage the transit without i_mutex
1188 if (managed & DCACHE_MANAGE_TRANSIT) {
1189 BUG_ON(!path->dentry->d_op);
1190 BUG_ON(!path->dentry->d_op->d_manage);
1191 ret = path->dentry->d_op->d_manage(path, false);
1196 /* Transit to a mounted filesystem. */
1197 if (managed & DCACHE_MOUNTED) {
1198 struct vfsmount *mounted = lookup_mnt(path);
1203 path->mnt = mounted;
1204 path->dentry = dget(mounted->mnt_root);
1209 /* Something is mounted on this dentry in another
1210 * namespace and/or whatever was mounted there in this
1211 * namespace got unmounted before lookup_mnt() could
1215 /* Handle an automount point */
1216 if (managed & DCACHE_NEED_AUTOMOUNT) {
1217 ret = follow_automount(path, nd, &need_mntput);
1223 /* We didn't change the current path point */
1227 if (need_mntput && path->mnt == mnt)
1229 if (ret == -EISDIR || !ret)
1232 nd->flags |= LOOKUP_JUMPED;
1233 if (unlikely(ret < 0))
1234 path_put_conditional(path, nd);
1238 int follow_down_one(struct path *path)
1240 struct vfsmount *mounted;
1242 mounted = lookup_mnt(path);
1246 path->mnt = mounted;
1247 path->dentry = dget(mounted->mnt_root);
1252 EXPORT_SYMBOL(follow_down_one);
1254 static inline int managed_dentry_rcu(const struct path *path)
1256 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1257 path->dentry->d_op->d_manage(path, true) : 0;
1261 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1262 * we meet a managed dentry that would need blocking.
1264 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1265 struct inode **inode, unsigned *seqp)
1268 struct mount *mounted;
1270 * Don't forget we might have a non-mountpoint managed dentry
1271 * that wants to block transit.
1273 switch (managed_dentry_rcu(path)) {
1283 if (!d_mountpoint(path->dentry))
1284 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1286 mounted = __lookup_mnt(path->mnt, path->dentry);
1289 path->mnt = &mounted->mnt;
1290 path->dentry = mounted->mnt.mnt_root;
1291 nd->flags |= LOOKUP_JUMPED;
1292 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1294 * Update the inode too. We don't need to re-check the
1295 * dentry sequence number here after this d_inode read,
1296 * because a mount-point is always pinned.
1298 *inode = path->dentry->d_inode;
1300 return !read_seqretry(&mount_lock, nd->m_seq) &&
1301 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1304 static int follow_dotdot_rcu(struct nameidata *nd)
1306 struct inode *inode = nd->inode;
1309 if (path_equal(&nd->path, &nd->root))
1311 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1312 struct dentry *old = nd->path.dentry;
1313 struct dentry *parent = old->d_parent;
1316 inode = parent->d_inode;
1317 seq = read_seqcount_begin(&parent->d_seq);
1318 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1320 nd->path.dentry = parent;
1322 if (unlikely(!path_connected(&nd->path)))
1326 struct mount *mnt = real_mount(nd->path.mnt);
1327 struct mount *mparent = mnt->mnt_parent;
1328 struct dentry *mountpoint = mnt->mnt_mountpoint;
1329 struct inode *inode2 = mountpoint->d_inode;
1330 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1331 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1333 if (&mparent->mnt == nd->path.mnt)
1335 /* we know that mountpoint was pinned */
1336 nd->path.dentry = mountpoint;
1337 nd->path.mnt = &mparent->mnt;
1342 while (unlikely(d_mountpoint(nd->path.dentry))) {
1343 struct mount *mounted;
1344 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1345 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1349 nd->path.mnt = &mounted->mnt;
1350 nd->path.dentry = mounted->mnt.mnt_root;
1351 inode = nd->path.dentry->d_inode;
1352 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1359 * Follow down to the covering mount currently visible to userspace. At each
1360 * point, the filesystem owning that dentry may be queried as to whether the
1361 * caller is permitted to proceed or not.
1363 int follow_down(struct path *path)
1368 while (managed = READ_ONCE(path->dentry->d_flags),
1369 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1370 /* Allow the filesystem to manage the transit without i_mutex
1373 * We indicate to the filesystem if someone is trying to mount
1374 * something here. This gives autofs the chance to deny anyone
1375 * other than its daemon the right to mount on its
1378 * The filesystem may sleep at this point.
1380 if (managed & DCACHE_MANAGE_TRANSIT) {
1381 BUG_ON(!path->dentry->d_op);
1382 BUG_ON(!path->dentry->d_op->d_manage);
1383 ret = path->dentry->d_op->d_manage(path, false);
1385 return ret == -EISDIR ? 0 : ret;
1388 /* Transit to a mounted filesystem. */
1389 if (managed & DCACHE_MOUNTED) {
1390 struct vfsmount *mounted = lookup_mnt(path);
1395 path->mnt = mounted;
1396 path->dentry = dget(mounted->mnt_root);
1400 /* Don't handle automount points here */
1405 EXPORT_SYMBOL(follow_down);
1408 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1410 static void follow_mount(struct path *path)
1412 while (d_mountpoint(path->dentry)) {
1413 struct vfsmount *mounted = lookup_mnt(path);
1418 path->mnt = mounted;
1419 path->dentry = dget(mounted->mnt_root);
1423 static int path_parent_directory(struct path *path)
1425 struct dentry *old = path->dentry;
1426 /* rare case of legitimate dget_parent()... */
1427 path->dentry = dget_parent(path->dentry);
1429 if (unlikely(!path_connected(path)))
1434 static int follow_dotdot(struct nameidata *nd)
1437 if (nd->path.dentry == nd->root.dentry &&
1438 nd->path.mnt == nd->root.mnt) {
1441 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1442 int ret = path_parent_directory(&nd->path);
1447 if (!follow_up(&nd->path))
1450 follow_mount(&nd->path);
1451 nd->inode = nd->path.dentry->d_inode;
1456 * This looks up the name in dcache and possibly revalidates the found dentry.
1457 * NULL is returned if the dentry does not exist in the cache.
1459 static struct dentry *lookup_dcache(const struct qstr *name,
1463 struct dentry *dentry = d_lookup(dir, name);
1465 int error = d_revalidate(dentry, flags);
1466 if (unlikely(error <= 0)) {
1468 d_invalidate(dentry);
1470 return ERR_PTR(error);
1477 * Parent directory has inode locked exclusive. This is one
1478 * and only case when ->lookup() gets called on non in-lookup
1479 * dentries - as the matter of fact, this only gets called
1480 * when directory is guaranteed to have no in-lookup children
1483 static struct dentry *__lookup_hash(const struct qstr *name,
1484 struct dentry *base, unsigned int flags)
1486 struct dentry *dentry = lookup_dcache(name, base, flags);
1488 struct inode *dir = base->d_inode;
1493 /* Don't create child dentry for a dead directory. */
1494 if (unlikely(IS_DEADDIR(dir)))
1495 return ERR_PTR(-ENOENT);
1497 dentry = d_alloc(base, name);
1498 if (unlikely(!dentry))
1499 return ERR_PTR(-ENOMEM);
1501 old = dir->i_op->lookup(dir, dentry, flags);
1502 if (unlikely(old)) {
1509 static int lookup_fast(struct nameidata *nd,
1510 struct path *path, struct inode **inode,
1513 struct vfsmount *mnt = nd->path.mnt;
1514 struct dentry *dentry, *parent = nd->path.dentry;
1519 * Rename seqlock is not required here because in the off chance
1520 * of a false negative due to a concurrent rename, the caller is
1521 * going to fall back to non-racy lookup.
1523 if (nd->flags & LOOKUP_RCU) {
1526 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1527 if (unlikely(!dentry)) {
1528 if (unlazy_walk(nd))
1534 * This sequence count validates that the inode matches
1535 * the dentry name information from lookup.
1537 *inode = d_backing_inode(dentry);
1538 negative = d_is_negative(dentry);
1539 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1543 * This sequence count validates that the parent had no
1544 * changes while we did the lookup of the dentry above.
1546 * The memory barrier in read_seqcount_begin of child is
1547 * enough, we can use __read_seqcount_retry here.
1549 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1553 status = d_revalidate(dentry, nd->flags);
1554 if (likely(status > 0)) {
1556 * Note: do negative dentry check after revalidation in
1557 * case that drops it.
1559 if (unlikely(negative))
1562 path->dentry = dentry;
1563 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1566 if (unlazy_child(nd, dentry, seq))
1568 if (unlikely(status == -ECHILD))
1569 /* we'd been told to redo it in non-rcu mode */
1570 status = d_revalidate(dentry, nd->flags);
1572 dentry = __d_lookup(parent, &nd->last);
1573 if (unlikely(!dentry))
1575 status = d_revalidate(dentry, nd->flags);
1577 if (unlikely(status <= 0)) {
1579 d_invalidate(dentry);
1583 if (unlikely(d_is_negative(dentry))) {
1589 path->dentry = dentry;
1590 err = follow_managed(path, nd);
1591 if (likely(err > 0))
1592 *inode = d_backing_inode(path->dentry);
1596 /* Fast lookup failed, do it the slow way */
1597 static struct dentry *lookup_slow(const struct qstr *name,
1601 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1602 struct inode *inode = dir->d_inode;
1603 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1605 inode_lock_shared(inode);
1606 /* Don't go there if it's already dead */
1607 if (unlikely(IS_DEADDIR(inode)))
1610 dentry = d_alloc_parallel(dir, name, &wq);
1613 if (unlikely(!d_in_lookup(dentry))) {
1614 if (!(flags & LOOKUP_NO_REVAL)) {
1615 int error = d_revalidate(dentry, flags);
1616 if (unlikely(error <= 0)) {
1618 d_invalidate(dentry);
1623 dentry = ERR_PTR(error);
1627 old = inode->i_op->lookup(inode, dentry, flags);
1628 d_lookup_done(dentry);
1629 if (unlikely(old)) {
1635 inode_unlock_shared(inode);
1639 static inline int may_lookup(struct nameidata *nd)
1641 if (nd->flags & LOOKUP_RCU) {
1642 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1645 if (unlazy_walk(nd))
1648 return inode_permission(nd->inode, MAY_EXEC);
1651 static inline int handle_dots(struct nameidata *nd, int type)
1653 if (type == LAST_DOTDOT) {
1656 if (nd->flags & LOOKUP_RCU) {
1657 return follow_dotdot_rcu(nd);
1659 return follow_dotdot(nd);
1664 static int pick_link(struct nameidata *nd, struct path *link,
1665 struct inode *inode, unsigned seq)
1669 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1670 path_to_nameidata(link, nd);
1673 if (!(nd->flags & LOOKUP_RCU)) {
1674 if (link->mnt == nd->path.mnt)
1677 error = nd_alloc_stack(nd);
1678 if (unlikely(error)) {
1679 if (error == -ECHILD) {
1680 if (unlikely(!legitimize_path(nd, link, seq))) {
1683 nd->flags &= ~LOOKUP_RCU;
1684 nd->path.mnt = NULL;
1685 nd->path.dentry = NULL;
1686 if (!(nd->flags & LOOKUP_ROOT))
1687 nd->root.mnt = NULL;
1689 } else if (likely(unlazy_walk(nd)) == 0)
1690 error = nd_alloc_stack(nd);
1698 last = nd->stack + nd->depth++;
1700 clear_delayed_call(&last->done);
1701 nd->link_inode = inode;
1706 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1709 * Do we need to follow links? We _really_ want to be able
1710 * to do this check without having to look at inode->i_op,
1711 * so we keep a cache of "no, this doesn't need follow_link"
1712 * for the common case.
1714 static inline int step_into(struct nameidata *nd, struct path *path,
1715 int flags, struct inode *inode, unsigned seq)
1717 if (!(flags & WALK_MORE) && nd->depth)
1719 if (likely(!d_is_symlink(path->dentry)) ||
1720 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1721 /* not a symlink or should not follow */
1722 path_to_nameidata(path, nd);
1727 /* make sure that d_is_symlink above matches inode */
1728 if (nd->flags & LOOKUP_RCU) {
1729 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1732 return pick_link(nd, path, inode, seq);
1735 static int walk_component(struct nameidata *nd, int flags)
1738 struct inode *inode;
1742 * "." and ".." are special - ".." especially so because it has
1743 * to be able to know about the current root directory and
1744 * parent relationships.
1746 if (unlikely(nd->last_type != LAST_NORM)) {
1747 err = handle_dots(nd, nd->last_type);
1748 if (!(flags & WALK_MORE) && nd->depth)
1752 err = lookup_fast(nd, &path, &inode, &seq);
1753 if (unlikely(err <= 0)) {
1756 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1758 if (IS_ERR(path.dentry))
1759 return PTR_ERR(path.dentry);
1761 path.mnt = nd->path.mnt;
1762 err = follow_managed(&path, nd);
1763 if (unlikely(err < 0))
1766 if (unlikely(d_is_negative(path.dentry))) {
1767 path_to_nameidata(&path, nd);
1771 seq = 0; /* we are already out of RCU mode */
1772 inode = d_backing_inode(path.dentry);
1775 return step_into(nd, &path, flags, inode, seq);
1779 * We can do the critical dentry name comparison and hashing
1780 * operations one word at a time, but we are limited to:
1782 * - Architectures with fast unaligned word accesses. We could
1783 * do a "get_unaligned()" if this helps and is sufficiently
1786 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1787 * do not trap on the (extremely unlikely) case of a page
1788 * crossing operation.
1790 * - Furthermore, we need an efficient 64-bit compile for the
1791 * 64-bit case in order to generate the "number of bytes in
1792 * the final mask". Again, that could be replaced with a
1793 * efficient population count instruction or similar.
1795 #ifdef CONFIG_DCACHE_WORD_ACCESS
1797 #include <asm/word-at-a-time.h>
1801 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1803 #elif defined(CONFIG_64BIT)
1805 * Register pressure in the mixing function is an issue, particularly
1806 * on 32-bit x86, but almost any function requires one state value and
1807 * one temporary. Instead, use a function designed for two state values
1808 * and no temporaries.
1810 * This function cannot create a collision in only two iterations, so
1811 * we have two iterations to achieve avalanche. In those two iterations,
1812 * we have six layers of mixing, which is enough to spread one bit's
1813 * influence out to 2^6 = 64 state bits.
1815 * Rotate constants are scored by considering either 64 one-bit input
1816 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1817 * probability of that delta causing a change to each of the 128 output
1818 * bits, using a sample of random initial states.
1820 * The Shannon entropy of the computed probabilities is then summed
1821 * to produce a score. Ideally, any input change has a 50% chance of
1822 * toggling any given output bit.
1824 * Mixing scores (in bits) for (12,45):
1825 * Input delta: 1-bit 2-bit
1826 * 1 round: 713.3 42542.6
1827 * 2 rounds: 2753.7 140389.8
1828 * 3 rounds: 5954.1 233458.2
1829 * 4 rounds: 7862.6 256672.2
1830 * Perfect: 8192 258048
1831 * (64*128) (64*63/2 * 128)
1833 #define HASH_MIX(x, y, a) \
1835 y ^= x, x = rol64(x,12),\
1836 x += y, y = rol64(y,45),\
1840 * Fold two longs into one 32-bit hash value. This must be fast, but
1841 * latency isn't quite as critical, as there is a fair bit of additional
1842 * work done before the hash value is used.
1844 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1846 y ^= x * GOLDEN_RATIO_64;
1847 y *= GOLDEN_RATIO_64;
1851 #else /* 32-bit case */
1854 * Mixing scores (in bits) for (7,20):
1855 * Input delta: 1-bit 2-bit
1856 * 1 round: 330.3 9201.6
1857 * 2 rounds: 1246.4 25475.4
1858 * 3 rounds: 1907.1 31295.1
1859 * 4 rounds: 2042.3 31718.6
1860 * Perfect: 2048 31744
1861 * (32*64) (32*31/2 * 64)
1863 #define HASH_MIX(x, y, a) \
1865 y ^= x, x = rol32(x, 7),\
1866 x += y, y = rol32(y,20),\
1869 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1871 /* Use arch-optimized multiply if one exists */
1872 return __hash_32(y ^ __hash_32(x));
1878 * Return the hash of a string of known length. This is carfully
1879 * designed to match hash_name(), which is the more critical function.
1880 * In particular, we must end by hashing a final word containing 0..7
1881 * payload bytes, to match the way that hash_name() iterates until it
1882 * finds the delimiter after the name.
1884 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1886 unsigned long a, x = 0, y = (unsigned long)salt;
1891 a = load_unaligned_zeropad(name);
1892 if (len < sizeof(unsigned long))
1895 name += sizeof(unsigned long);
1896 len -= sizeof(unsigned long);
1898 x ^= a & bytemask_from_count(len);
1900 return fold_hash(x, y);
1902 EXPORT_SYMBOL(full_name_hash);
1904 /* Return the "hash_len" (hash and length) of a null-terminated string */
1905 u64 hashlen_string(const void *salt, const char *name)
1907 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1908 unsigned long adata, mask, len;
1909 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1916 len += sizeof(unsigned long);
1918 a = load_unaligned_zeropad(name+len);
1919 } while (!has_zero(a, &adata, &constants));
1921 adata = prep_zero_mask(a, adata, &constants);
1922 mask = create_zero_mask(adata);
1923 x ^= a & zero_bytemask(mask);
1925 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1927 EXPORT_SYMBOL(hashlen_string);
1930 * Calculate the length and hash of the path component, and
1931 * return the "hash_len" as the result.
1933 static inline u64 hash_name(const void *salt, const char *name)
1935 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1936 unsigned long adata, bdata, mask, len;
1937 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1944 len += sizeof(unsigned long);
1946 a = load_unaligned_zeropad(name+len);
1947 b = a ^ REPEAT_BYTE('/');
1948 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1950 adata = prep_zero_mask(a, adata, &constants);
1951 bdata = prep_zero_mask(b, bdata, &constants);
1952 mask = create_zero_mask(adata | bdata);
1953 x ^= a & zero_bytemask(mask);
1955 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1958 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1960 /* Return the hash of a string of known length */
1961 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1963 unsigned long hash = init_name_hash(salt);
1965 hash = partial_name_hash((unsigned char)*name++, hash);
1966 return end_name_hash(hash);
1968 EXPORT_SYMBOL(full_name_hash);
1970 /* Return the "hash_len" (hash and length) of a null-terminated string */
1971 u64 hashlen_string(const void *salt, const char *name)
1973 unsigned long hash = init_name_hash(salt);
1974 unsigned long len = 0, c;
1976 c = (unsigned char)*name;
1979 hash = partial_name_hash(c, hash);
1980 c = (unsigned char)name[len];
1982 return hashlen_create(end_name_hash(hash), len);
1984 EXPORT_SYMBOL(hashlen_string);
1987 * We know there's a real path component here of at least
1990 static inline u64 hash_name(const void *salt, const char *name)
1992 unsigned long hash = init_name_hash(salt);
1993 unsigned long len = 0, c;
1995 c = (unsigned char)*name;
1998 hash = partial_name_hash(c, hash);
1999 c = (unsigned char)name[len];
2000 } while (c && c != '/');
2001 return hashlen_create(end_name_hash(hash), len);
2008 * This is the basic name resolution function, turning a pathname into
2009 * the final dentry. We expect 'base' to be positive and a directory.
2011 * Returns 0 and nd will have valid dentry and mnt on success.
2012 * Returns error and drops reference to input namei data on failure.
2014 static int link_path_walk(const char *name, struct nameidata *nd)
2023 /* At this point we know we have a real path component. */
2028 err = may_lookup(nd);
2032 hash_len = hash_name(nd->path.dentry, name);
2035 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2037 if (name[1] == '.') {
2039 nd->flags |= LOOKUP_JUMPED;
2045 if (likely(type == LAST_NORM)) {
2046 struct dentry *parent = nd->path.dentry;
2047 nd->flags &= ~LOOKUP_JUMPED;
2048 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2049 struct qstr this = { { .hash_len = hash_len }, .name = name };
2050 err = parent->d_op->d_hash(parent, &this);
2053 hash_len = this.hash_len;
2058 nd->last.hash_len = hash_len;
2059 nd->last.name = name;
2060 nd->last_type = type;
2062 name += hashlen_len(hash_len);
2066 * If it wasn't NUL, we know it was '/'. Skip that
2067 * slash, and continue until no more slashes.
2071 } while (unlikely(*name == '/'));
2072 if (unlikely(!*name)) {
2074 /* pathname body, done */
2077 name = nd->stack[nd->depth - 1].name;
2078 /* trailing symlink, done */
2081 /* last component of nested symlink */
2082 err = walk_component(nd, WALK_FOLLOW);
2084 /* not the last component */
2085 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2091 const char *s = get_link(nd);
2100 nd->stack[nd->depth - 1].name = name;
2105 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2106 if (nd->flags & LOOKUP_RCU) {
2107 if (unlazy_walk(nd))
2115 static const char *path_init(struct nameidata *nd, unsigned flags)
2117 const char *s = nd->name->name;
2120 flags &= ~LOOKUP_RCU;
2122 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2123 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2125 if (flags & LOOKUP_ROOT) {
2126 struct dentry *root = nd->root.dentry;
2127 struct inode *inode = root->d_inode;
2128 if (*s && unlikely(!d_can_lookup(root)))
2129 return ERR_PTR(-ENOTDIR);
2130 nd->path = nd->root;
2132 if (flags & LOOKUP_RCU) {
2134 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2135 nd->root_seq = nd->seq;
2136 nd->m_seq = read_seqbegin(&mount_lock);
2138 path_get(&nd->path);
2143 nd->root.mnt = NULL;
2144 nd->path.mnt = NULL;
2145 nd->path.dentry = NULL;
2147 nd->m_seq = read_seqbegin(&mount_lock);
2149 if (flags & LOOKUP_RCU)
2152 if (likely(!nd_jump_root(nd)))
2154 nd->root.mnt = NULL;
2156 return ERR_PTR(-ECHILD);
2157 } else if (nd->dfd == AT_FDCWD) {
2158 if (flags & LOOKUP_RCU) {
2159 struct fs_struct *fs = current->fs;
2165 seq = read_seqcount_begin(&fs->seq);
2167 nd->inode = nd->path.dentry->d_inode;
2168 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2169 } while (read_seqcount_retry(&fs->seq, seq));
2171 get_fs_pwd(current->fs, &nd->path);
2172 nd->inode = nd->path.dentry->d_inode;
2176 /* Caller must check execute permissions on the starting path component */
2177 struct fd f = fdget_raw(nd->dfd);
2178 struct dentry *dentry;
2181 return ERR_PTR(-EBADF);
2183 dentry = f.file->f_path.dentry;
2186 if (!d_can_lookup(dentry)) {
2188 return ERR_PTR(-ENOTDIR);
2192 nd->path = f.file->f_path;
2193 if (flags & LOOKUP_RCU) {
2195 nd->inode = nd->path.dentry->d_inode;
2196 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2198 path_get(&nd->path);
2199 nd->inode = nd->path.dentry->d_inode;
2206 static const char *trailing_symlink(struct nameidata *nd)
2209 int error = may_follow_link(nd);
2210 if (unlikely(error))
2211 return ERR_PTR(error);
2212 nd->flags |= LOOKUP_PARENT;
2213 nd->stack[0].name = NULL;
2218 static inline int lookup_last(struct nameidata *nd)
2220 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2221 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2223 nd->flags &= ~LOOKUP_PARENT;
2224 return walk_component(nd, 0);
2227 static int handle_lookup_down(struct nameidata *nd)
2229 struct path path = nd->path;
2230 struct inode *inode = nd->inode;
2231 unsigned seq = nd->seq;
2234 if (nd->flags & LOOKUP_RCU) {
2236 * don't bother with unlazy_walk on failure - we are
2237 * at the very beginning of walk, so we lose nothing
2238 * if we simply redo everything in non-RCU mode
2240 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2244 err = follow_managed(&path, nd);
2245 if (unlikely(err < 0))
2247 inode = d_backing_inode(path.dentry);
2250 path_to_nameidata(&path, nd);
2256 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2257 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2259 const char *s = path_init(nd, flags);
2265 if (unlikely(flags & LOOKUP_DOWN)) {
2266 err = handle_lookup_down(nd);
2267 if (unlikely(err < 0)) {
2273 while (!(err = link_path_walk(s, nd))
2274 && ((err = lookup_last(nd)) > 0)) {
2275 s = trailing_symlink(nd);
2282 err = complete_walk(nd);
2284 if (!err && nd->flags & LOOKUP_DIRECTORY)
2285 if (!d_can_lookup(nd->path.dentry))
2289 nd->path.mnt = NULL;
2290 nd->path.dentry = NULL;
2296 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2297 struct path *path, struct path *root)
2300 struct nameidata nd;
2302 return PTR_ERR(name);
2303 if (unlikely(root)) {
2305 flags |= LOOKUP_ROOT;
2307 set_nameidata(&nd, dfd, name);
2308 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2309 if (unlikely(retval == -ECHILD))
2310 retval = path_lookupat(&nd, flags, path);
2311 if (unlikely(retval == -ESTALE))
2312 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2314 if (likely(!retval))
2315 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2316 restore_nameidata();
2321 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2322 static int path_parentat(struct nameidata *nd, unsigned flags,
2323 struct path *parent)
2325 const char *s = path_init(nd, flags);
2329 err = link_path_walk(s, nd);
2331 err = complete_walk(nd);
2334 nd->path.mnt = NULL;
2335 nd->path.dentry = NULL;
2341 static struct filename *filename_parentat(int dfd, struct filename *name,
2342 unsigned int flags, struct path *parent,
2343 struct qstr *last, int *type)
2346 struct nameidata nd;
2350 set_nameidata(&nd, dfd, name);
2351 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2352 if (unlikely(retval == -ECHILD))
2353 retval = path_parentat(&nd, flags, parent);
2354 if (unlikely(retval == -ESTALE))
2355 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2356 if (likely(!retval)) {
2358 *type = nd.last_type;
2359 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2362 name = ERR_PTR(retval);
2364 restore_nameidata();
2368 /* does lookup, returns the object with parent locked */
2369 struct dentry *kern_path_locked(const char *name, struct path *path)
2371 struct filename *filename;
2376 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2378 if (IS_ERR(filename))
2379 return ERR_CAST(filename);
2380 if (unlikely(type != LAST_NORM)) {
2383 return ERR_PTR(-EINVAL);
2385 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2386 d = __lookup_hash(&last, path->dentry, 0);
2388 inode_unlock(path->dentry->d_inode);
2395 int kern_path(const char *name, unsigned int flags, struct path *path)
2397 return filename_lookup(AT_FDCWD, getname_kernel(name),
2400 EXPORT_SYMBOL(kern_path);
2403 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2404 * @dentry: pointer to dentry of the base directory
2405 * @mnt: pointer to vfs mount of the base directory
2406 * @name: pointer to file name
2407 * @flags: lookup flags
2408 * @path: pointer to struct path to fill
2410 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2411 const char *name, unsigned int flags,
2414 struct path root = {.mnt = mnt, .dentry = dentry};
2415 /* the first argument of filename_lookup() is ignored with root */
2416 return filename_lookup(AT_FDCWD, getname_kernel(name),
2417 flags , path, &root);
2419 EXPORT_SYMBOL(vfs_path_lookup);
2422 * lookup_one_len - filesystem helper to lookup single pathname component
2423 * @name: pathname component to lookup
2424 * @base: base directory to lookup from
2425 * @len: maximum length @len should be interpreted to
2427 * Note that this routine is purely a helper for filesystem usage and should
2428 * not be called by generic code.
2430 * The caller must hold base->i_mutex.
2432 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2438 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2442 this.hash = full_name_hash(base, name, len);
2444 return ERR_PTR(-EACCES);
2446 if (unlikely(name[0] == '.')) {
2447 if (len < 2 || (len == 2 && name[1] == '.'))
2448 return ERR_PTR(-EACCES);
2452 c = *(const unsigned char *)name++;
2453 if (c == '/' || c == '\0')
2454 return ERR_PTR(-EACCES);
2457 * See if the low-level filesystem might want
2458 * to use its own hash..
2460 if (base->d_flags & DCACHE_OP_HASH) {
2461 int err = base->d_op->d_hash(base, &this);
2463 return ERR_PTR(err);
2466 err = inode_permission(base->d_inode, MAY_EXEC);
2468 return ERR_PTR(err);
2470 return __lookup_hash(&this, base, 0);
2472 EXPORT_SYMBOL(lookup_one_len);
2475 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2476 * @name: pathname component to lookup
2477 * @base: base directory to lookup from
2478 * @len: maximum length @len should be interpreted to
2480 * Note that this routine is purely a helper for filesystem usage and should
2481 * not be called by generic code.
2483 * Unlike lookup_one_len, it should be called without the parent
2484 * i_mutex held, and will take the i_mutex itself if necessary.
2486 struct dentry *lookup_one_len_unlocked(const char *name,
2487 struct dentry *base, int len)
2496 this.hash = full_name_hash(base, name, len);
2498 return ERR_PTR(-EACCES);
2500 if (unlikely(name[0] == '.')) {
2501 if (len < 2 || (len == 2 && name[1] == '.'))
2502 return ERR_PTR(-EACCES);
2506 c = *(const unsigned char *)name++;
2507 if (c == '/' || c == '\0')
2508 return ERR_PTR(-EACCES);
2511 * See if the low-level filesystem might want
2512 * to use its own hash..
2514 if (base->d_flags & DCACHE_OP_HASH) {
2515 int err = base->d_op->d_hash(base, &this);
2517 return ERR_PTR(err);
2520 err = inode_permission(base->d_inode, MAY_EXEC);
2522 return ERR_PTR(err);
2524 ret = lookup_dcache(&this, base, 0);
2526 ret = lookup_slow(&this, base, 0);
2529 EXPORT_SYMBOL(lookup_one_len_unlocked);
2531 #ifdef CONFIG_UNIX98_PTYS
2532 int path_pts(struct path *path)
2534 /* Find something mounted on "pts" in the same directory as
2537 struct dentry *child, *parent;
2541 ret = path_parent_directory(path);
2545 parent = path->dentry;
2548 child = d_hash_and_lookup(parent, &this);
2552 path->dentry = child;
2559 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2560 struct path *path, int *empty)
2562 return filename_lookup(dfd, getname_flags(name, flags, empty),
2565 EXPORT_SYMBOL(user_path_at_empty);
2568 * mountpoint_last - look up last component for umount
2569 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2571 * This is a special lookup_last function just for umount. In this case, we
2572 * need to resolve the path without doing any revalidation.
2574 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2575 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2576 * in almost all cases, this lookup will be served out of the dcache. The only
2577 * cases where it won't are if nd->last refers to a symlink or the path is
2578 * bogus and it doesn't exist.
2581 * -error: if there was an error during lookup. This includes -ENOENT if the
2582 * lookup found a negative dentry.
2584 * 0: if we successfully resolved nd->last and found it to not to be a
2585 * symlink that needs to be followed.
2587 * 1: if we successfully resolved nd->last and found it to be a symlink
2588 * that needs to be followed.
2591 mountpoint_last(struct nameidata *nd)
2594 struct dentry *dir = nd->path.dentry;
2597 /* If we're in rcuwalk, drop out of it to handle last component */
2598 if (nd->flags & LOOKUP_RCU) {
2599 if (unlazy_walk(nd))
2603 nd->flags &= ~LOOKUP_PARENT;
2605 if (unlikely(nd->last_type != LAST_NORM)) {
2606 error = handle_dots(nd, nd->last_type);
2609 path.dentry = dget(nd->path.dentry);
2611 path.dentry = d_lookup(dir, &nd->last);
2614 * No cached dentry. Mounted dentries are pinned in the
2615 * cache, so that means that this dentry is probably
2616 * a symlink or the path doesn't actually point
2617 * to a mounted dentry.
2619 path.dentry = lookup_slow(&nd->last, dir,
2620 nd->flags | LOOKUP_NO_REVAL);
2621 if (IS_ERR(path.dentry))
2622 return PTR_ERR(path.dentry);
2625 if (d_is_negative(path.dentry)) {
2629 path.mnt = nd->path.mnt;
2630 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2634 * path_mountpoint - look up a path to be umounted
2635 * @nd: lookup context
2636 * @flags: lookup flags
2637 * @path: pointer to container for result
2639 * Look up the given name, but don't attempt to revalidate the last component.
2640 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2643 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2645 const char *s = path_init(nd, flags);
2649 while (!(err = link_path_walk(s, nd)) &&
2650 (err = mountpoint_last(nd)) > 0) {
2651 s = trailing_symlink(nd);
2659 nd->path.mnt = NULL;
2660 nd->path.dentry = NULL;
2668 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2671 struct nameidata nd;
2674 return PTR_ERR(name);
2675 set_nameidata(&nd, dfd, name);
2676 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2677 if (unlikely(error == -ECHILD))
2678 error = path_mountpoint(&nd, flags, path);
2679 if (unlikely(error == -ESTALE))
2680 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2682 audit_inode(name, path->dentry, 0);
2683 restore_nameidata();
2689 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2690 * @dfd: directory file descriptor
2691 * @name: pathname from userland
2692 * @flags: lookup flags
2693 * @path: pointer to container to hold result
2695 * A umount is a special case for path walking. We're not actually interested
2696 * in the inode in this situation, and ESTALE errors can be a problem. We
2697 * simply want track down the dentry and vfsmount attached at the mountpoint
2698 * and avoid revalidating the last component.
2700 * Returns 0 and populates "path" on success.
2703 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2706 return filename_mountpoint(dfd, getname(name), path, flags);
2710 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2713 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2715 EXPORT_SYMBOL(kern_path_mountpoint);
2717 int __check_sticky(struct inode *dir, struct inode *inode)
2719 kuid_t fsuid = current_fsuid();
2721 if (uid_eq(inode->i_uid, fsuid))
2723 if (uid_eq(dir->i_uid, fsuid))
2725 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2727 EXPORT_SYMBOL(__check_sticky);
2730 * Check whether we can remove a link victim from directory dir, check
2731 * whether the type of victim is right.
2732 * 1. We can't do it if dir is read-only (done in permission())
2733 * 2. We should have write and exec permissions on dir
2734 * 3. We can't remove anything from append-only dir
2735 * 4. We can't do anything with immutable dir (done in permission())
2736 * 5. If the sticky bit on dir is set we should either
2737 * a. be owner of dir, or
2738 * b. be owner of victim, or
2739 * c. have CAP_FOWNER capability
2740 * 6. If the victim is append-only or immutable we can't do antyhing with
2741 * links pointing to it.
2742 * 7. If the victim has an unknown uid or gid we can't change the inode.
2743 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2744 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2745 * 10. We can't remove a root or mountpoint.
2746 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2747 * nfs_async_unlink().
2749 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2751 struct inode *inode = d_backing_inode(victim);
2754 if (d_is_negative(victim))
2758 BUG_ON(victim->d_parent->d_inode != dir);
2759 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2761 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2767 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2768 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2771 if (!d_is_dir(victim))
2773 if (IS_ROOT(victim))
2775 } else if (d_is_dir(victim))
2777 if (IS_DEADDIR(dir))
2779 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2784 /* Check whether we can create an object with dentry child in directory
2786 * 1. We can't do it if child already exists (open has special treatment for
2787 * this case, but since we are inlined it's OK)
2788 * 2. We can't do it if dir is read-only (done in permission())
2789 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2790 * 4. We should have write and exec permissions on dir
2791 * 5. We can't do it if dir is immutable (done in permission())
2793 static inline int may_create(struct inode *dir, struct dentry *child)
2795 struct user_namespace *s_user_ns;
2796 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2799 if (IS_DEADDIR(dir))
2801 s_user_ns = dir->i_sb->s_user_ns;
2802 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2803 !kgid_has_mapping(s_user_ns, current_fsgid()))
2805 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2809 * p1 and p2 should be directories on the same fs.
2811 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2816 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2820 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2822 p = d_ancestor(p2, p1);
2824 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2825 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2829 p = d_ancestor(p1, p2);
2831 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2832 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2836 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2837 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2840 EXPORT_SYMBOL(lock_rename);
2842 void unlock_rename(struct dentry *p1, struct dentry *p2)
2844 inode_unlock(p1->d_inode);
2846 inode_unlock(p2->d_inode);
2847 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2850 EXPORT_SYMBOL(unlock_rename);
2852 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2855 int error = may_create(dir, dentry);
2859 if (!dir->i_op->create)
2860 return -EACCES; /* shouldn't it be ENOSYS? */
2863 error = security_inode_create(dir, dentry, mode);
2866 error = dir->i_op->create(dir, dentry, mode, want_excl);
2868 fsnotify_create(dir, dentry);
2871 EXPORT_SYMBOL(vfs_create);
2873 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2874 int (*f)(struct dentry *, umode_t, void *),
2877 struct inode *dir = dentry->d_parent->d_inode;
2878 int error = may_create(dir, dentry);
2884 error = security_inode_create(dir, dentry, mode);
2887 error = f(dentry, mode, arg);
2889 fsnotify_create(dir, dentry);
2892 EXPORT_SYMBOL(vfs_mkobj);
2894 bool may_open_dev(const struct path *path)
2896 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2897 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2900 static int may_open(const struct path *path, int acc_mode, int flag)
2902 struct dentry *dentry = path->dentry;
2903 struct inode *inode = dentry->d_inode;
2909 switch (inode->i_mode & S_IFMT) {
2913 if (acc_mode & MAY_WRITE)
2918 if (!may_open_dev(path))
2927 error = inode_permission(inode, MAY_OPEN | acc_mode);
2932 * An append-only file must be opened in append mode for writing.
2934 if (IS_APPEND(inode)) {
2935 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2941 /* O_NOATIME can only be set by the owner or superuser */
2942 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2948 static int handle_truncate(struct file *filp)
2950 const struct path *path = &filp->f_path;
2951 struct inode *inode = path->dentry->d_inode;
2952 int error = get_write_access(inode);
2956 * Refuse to truncate files with mandatory locks held on them.
2958 error = locks_verify_locked(filp);
2960 error = security_path_truncate(path);
2962 error = do_truncate(path->dentry, 0,
2963 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2966 put_write_access(inode);
2970 static inline int open_to_namei_flags(int flag)
2972 if ((flag & O_ACCMODE) == 3)
2977 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2979 struct user_namespace *s_user_ns;
2980 int error = security_path_mknod(dir, dentry, mode, 0);
2984 s_user_ns = dir->dentry->d_sb->s_user_ns;
2985 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2986 !kgid_has_mapping(s_user_ns, current_fsgid()))
2989 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2993 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2997 * Attempt to atomically look up, create and open a file from a negative
3000 * Returns 0 if successful. The file will have been created and attached to
3001 * @file by the filesystem calling finish_open().
3003 * Returns 1 if the file was looked up only or didn't need creating. The
3004 * caller will need to perform the open themselves. @path will have been
3005 * updated to point to the new dentry. This may be negative.
3007 * Returns an error code otherwise.
3009 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3010 struct path *path, struct file *file,
3011 const struct open_flags *op,
3012 int open_flag, umode_t mode,
3015 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3016 struct inode *dir = nd->path.dentry->d_inode;
3019 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3020 open_flag &= ~O_TRUNC;
3022 if (nd->flags & LOOKUP_DIRECTORY)
3023 open_flag |= O_DIRECTORY;
3025 file->f_path.dentry = DENTRY_NOT_SET;
3026 file->f_path.mnt = nd->path.mnt;
3027 error = dir->i_op->atomic_open(dir, dentry, file,
3028 open_to_namei_flags(open_flag),
3030 d_lookup_done(dentry);
3033 * We didn't have the inode before the open, so check open
3036 int acc_mode = op->acc_mode;
3037 if (*opened & FILE_CREATED) {
3038 WARN_ON(!(open_flag & O_CREAT));
3039 fsnotify_create(dir, dentry);
3042 error = may_open(&file->f_path, acc_mode, open_flag);
3043 if (WARN_ON(error > 0))
3045 } else if (error > 0) {
3046 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3049 if (file->f_path.dentry) {
3051 dentry = file->f_path.dentry;
3053 if (*opened & FILE_CREATED)
3054 fsnotify_create(dir, dentry);
3055 if (unlikely(d_is_negative(dentry))) {
3058 path->dentry = dentry;
3059 path->mnt = nd->path.mnt;
3069 * Look up and maybe create and open the last component.
3071 * Must be called with i_mutex held on parent.
3073 * Returns 0 if the file was successfully atomically created (if necessary) and
3074 * opened. In this case the file will be returned attached to @file.
3076 * Returns 1 if the file was not completely opened at this time, though lookups
3077 * and creations will have been performed and the dentry returned in @path will
3078 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3079 * specified then a negative dentry may be returned.
3081 * An error code is returned otherwise.
3083 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3084 * cleared otherwise prior to returning.
3086 static int lookup_open(struct nameidata *nd, struct path *path,
3088 const struct open_flags *op,
3089 bool got_write, int *opened)
3091 struct dentry *dir = nd->path.dentry;
3092 struct inode *dir_inode = dir->d_inode;
3093 int open_flag = op->open_flag;
3094 struct dentry *dentry;
3095 int error, create_error = 0;
3096 umode_t mode = op->mode;
3097 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3099 if (unlikely(IS_DEADDIR(dir_inode)))
3102 *opened &= ~FILE_CREATED;
3103 dentry = d_lookup(dir, &nd->last);
3106 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3108 return PTR_ERR(dentry);
3110 if (d_in_lookup(dentry))
3113 error = d_revalidate(dentry, nd->flags);
3114 if (likely(error > 0))
3118 d_invalidate(dentry);
3122 if (dentry->d_inode) {
3123 /* Cached positive dentry: will open in f_op->open */
3128 * Checking write permission is tricky, bacuse we don't know if we are
3129 * going to actually need it: O_CREAT opens should work as long as the
3130 * file exists. But checking existence breaks atomicity. The trick is
3131 * to check access and if not granted clear O_CREAT from the flags.
3133 * Another problem is returing the "right" error value (e.g. for an
3134 * O_EXCL open we want to return EEXIST not EROFS).
3136 if (open_flag & O_CREAT) {
3137 if (!IS_POSIXACL(dir->d_inode))
3138 mode &= ~current_umask();
3139 if (unlikely(!got_write)) {
3140 create_error = -EROFS;
3141 open_flag &= ~O_CREAT;
3142 if (open_flag & (O_EXCL | O_TRUNC))
3144 /* No side effects, safe to clear O_CREAT */
3146 create_error = may_o_create(&nd->path, dentry, mode);
3148 open_flag &= ~O_CREAT;
3149 if (open_flag & O_EXCL)
3153 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3154 unlikely(!got_write)) {
3156 * No O_CREATE -> atomicity not a requirement -> fall
3157 * back to lookup + open
3162 if (dir_inode->i_op->atomic_open) {
3163 error = atomic_open(nd, dentry, path, file, op, open_flag,
3165 if (unlikely(error == -ENOENT) && create_error)
3166 error = create_error;
3171 if (d_in_lookup(dentry)) {
3172 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3174 d_lookup_done(dentry);
3175 if (unlikely(res)) {
3177 error = PTR_ERR(res);
3185 /* Negative dentry, just create the file */
3186 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3187 *opened |= FILE_CREATED;
3188 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3189 if (!dir_inode->i_op->create) {
3193 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3194 open_flag & O_EXCL);
3197 fsnotify_create(dir_inode, dentry);
3199 if (unlikely(create_error) && !dentry->d_inode) {
3200 error = create_error;
3204 path->dentry = dentry;
3205 path->mnt = nd->path.mnt;
3214 * Handle the last step of open()
3216 static int do_last(struct nameidata *nd,
3217 struct file *file, const struct open_flags *op,
3220 struct dentry *dir = nd->path.dentry;
3221 int open_flag = op->open_flag;
3222 bool will_truncate = (open_flag & O_TRUNC) != 0;
3223 bool got_write = false;
3224 int acc_mode = op->acc_mode;
3226 struct inode *inode;
3230 nd->flags &= ~LOOKUP_PARENT;
3231 nd->flags |= op->intent;
3233 if (nd->last_type != LAST_NORM) {
3234 error = handle_dots(nd, nd->last_type);
3235 if (unlikely(error))
3240 if (!(open_flag & O_CREAT)) {
3241 if (nd->last.name[nd->last.len])
3242 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3243 /* we _can_ be in RCU mode here */
3244 error = lookup_fast(nd, &path, &inode, &seq);
3245 if (likely(error > 0))
3251 BUG_ON(nd->inode != dir->d_inode);
3252 BUG_ON(nd->flags & LOOKUP_RCU);
3254 /* create side of things */
3256 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3257 * has been cleared when we got to the last component we are
3260 error = complete_walk(nd);
3264 audit_inode(nd->name, dir, LOOKUP_PARENT);
3265 /* trailing slashes? */
3266 if (unlikely(nd->last.name[nd->last.len]))
3270 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3271 error = mnt_want_write(nd->path.mnt);
3275 * do _not_ fail yet - we might not need that or fail with
3276 * a different error; let lookup_open() decide; we'll be
3277 * dropping this one anyway.
3280 if (open_flag & O_CREAT)
3281 inode_lock(dir->d_inode);
3283 inode_lock_shared(dir->d_inode);
3284 error = lookup_open(nd, &path, file, op, got_write, opened);
3285 if (open_flag & O_CREAT)
3286 inode_unlock(dir->d_inode);
3288 inode_unlock_shared(dir->d_inode);
3294 if ((*opened & FILE_CREATED) ||
3295 !S_ISREG(file_inode(file)->i_mode))
3296 will_truncate = false;
3298 audit_inode(nd->name, file->f_path.dentry, 0);
3302 if (*opened & FILE_CREATED) {
3303 /* Don't check for write permission, don't truncate */
3304 open_flag &= ~O_TRUNC;
3305 will_truncate = false;
3307 path_to_nameidata(&path, nd);
3308 goto finish_open_created;
3312 * If atomic_open() acquired write access it is dropped now due to
3313 * possible mount and symlink following (this might be optimized away if
3317 mnt_drop_write(nd->path.mnt);
3321 error = follow_managed(&path, nd);
3322 if (unlikely(error < 0))
3325 if (unlikely(d_is_negative(path.dentry))) {
3326 path_to_nameidata(&path, nd);
3331 * create/update audit record if it already exists.
3333 audit_inode(nd->name, path.dentry, 0);
3335 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3336 path_to_nameidata(&path, nd);
3340 seq = 0; /* out of RCU mode, so the value doesn't matter */
3341 inode = d_backing_inode(path.dentry);
3343 error = step_into(nd, &path, 0, inode, seq);
3344 if (unlikely(error))
3347 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3348 error = complete_walk(nd);
3351 audit_inode(nd->name, nd->path.dentry, 0);
3353 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3356 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3358 if (!d_is_reg(nd->path.dentry))
3359 will_truncate = false;
3361 if (will_truncate) {
3362 error = mnt_want_write(nd->path.mnt);
3367 finish_open_created:
3368 error = may_open(&nd->path, acc_mode, open_flag);
3371 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3372 error = vfs_open(&nd->path, file, current_cred());
3375 *opened |= FILE_OPENED;
3377 error = open_check_o_direct(file);
3379 error = ima_file_check(file, op->acc_mode, *opened);
3380 if (!error && will_truncate)
3381 error = handle_truncate(file);
3383 if (unlikely(error) && (*opened & FILE_OPENED))
3385 if (unlikely(error > 0)) {
3390 mnt_drop_write(nd->path.mnt);
3394 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3396 struct dentry *child = NULL;
3397 struct inode *dir = dentry->d_inode;
3398 struct inode *inode;
3401 /* we want directory to be writable */
3402 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3405 error = -EOPNOTSUPP;
3406 if (!dir->i_op->tmpfile)
3409 child = d_alloc(dentry, &slash_name);
3410 if (unlikely(!child))
3412 error = dir->i_op->tmpfile(dir, child, mode);
3416 inode = child->d_inode;
3417 if (unlikely(!inode))
3419 if (!(open_flag & O_EXCL)) {
3420 spin_lock(&inode->i_lock);
3421 inode->i_state |= I_LINKABLE;
3422 spin_unlock(&inode->i_lock);
3428 return ERR_PTR(error);
3430 EXPORT_SYMBOL(vfs_tmpfile);
3432 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3433 const struct open_flags *op,
3434 struct file *file, int *opened)
3436 struct dentry *child;
3438 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3439 if (unlikely(error))
3441 error = mnt_want_write(path.mnt);
3442 if (unlikely(error))
3444 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3445 error = PTR_ERR(child);
3449 path.dentry = child;
3450 audit_inode(nd->name, child, 0);
3451 /* Don't check for other permissions, the inode was just created */
3452 error = may_open(&path, 0, op->open_flag);
3455 file->f_path.mnt = path.mnt;
3456 error = finish_open(file, child, NULL, opened);
3459 error = open_check_o_direct(file);
3463 mnt_drop_write(path.mnt);
3469 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3472 int error = path_lookupat(nd, flags, &path);
3474 audit_inode(nd->name, path.dentry, 0);
3475 error = vfs_open(&path, file, current_cred());
3481 static struct file *path_openat(struct nameidata *nd,
3482 const struct open_flags *op, unsigned flags)
3489 file = get_empty_filp();
3493 file->f_flags = op->open_flag;
3495 if (unlikely(file->f_flags & __O_TMPFILE)) {
3496 error = do_tmpfile(nd, flags, op, file, &opened);
3500 if (unlikely(file->f_flags & O_PATH)) {
3501 error = do_o_path(nd, flags, file);
3503 opened |= FILE_OPENED;
3507 s = path_init(nd, flags);
3512 while (!(error = link_path_walk(s, nd)) &&
3513 (error = do_last(nd, file, op, &opened)) > 0) {
3514 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3515 s = trailing_symlink(nd);
3523 if (!(opened & FILE_OPENED)) {
3527 if (unlikely(error)) {
3528 if (error == -EOPENSTALE) {
3529 if (flags & LOOKUP_RCU)
3534 file = ERR_PTR(error);
3539 struct file *do_filp_open(int dfd, struct filename *pathname,
3540 const struct open_flags *op)
3542 struct nameidata nd;
3543 int flags = op->lookup_flags;
3546 set_nameidata(&nd, dfd, pathname);
3547 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3548 if (unlikely(filp == ERR_PTR(-ECHILD)))
3549 filp = path_openat(&nd, op, flags);
3550 if (unlikely(filp == ERR_PTR(-ESTALE)))
3551 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3552 restore_nameidata();
3556 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3557 const char *name, const struct open_flags *op)
3559 struct nameidata nd;
3561 struct filename *filename;
3562 int flags = op->lookup_flags | LOOKUP_ROOT;
3565 nd.root.dentry = dentry;
3567 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3568 return ERR_PTR(-ELOOP);
3570 filename = getname_kernel(name);
3571 if (IS_ERR(filename))
3572 return ERR_CAST(filename);
3574 set_nameidata(&nd, -1, filename);
3575 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3576 if (unlikely(file == ERR_PTR(-ECHILD)))
3577 file = path_openat(&nd, op, flags);
3578 if (unlikely(file == ERR_PTR(-ESTALE)))
3579 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3580 restore_nameidata();
3585 static struct dentry *filename_create(int dfd, struct filename *name,
3586 struct path *path, unsigned int lookup_flags)
3588 struct dentry *dentry = ERR_PTR(-EEXIST);
3593 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3596 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3597 * other flags passed in are ignored!
3599 lookup_flags &= LOOKUP_REVAL;
3601 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3603 return ERR_CAST(name);
3606 * Yucky last component or no last component at all?
3607 * (foo/., foo/.., /////)
3609 if (unlikely(type != LAST_NORM))
3612 /* don't fail immediately if it's r/o, at least try to report other errors */
3613 err2 = mnt_want_write(path->mnt);
3615 * Do the final lookup.
3617 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3618 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3619 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3624 if (d_is_positive(dentry))
3628 * Special case - lookup gave negative, but... we had foo/bar/
3629 * From the vfs_mknod() POV we just have a negative dentry -
3630 * all is fine. Let's be bastards - you had / on the end, you've
3631 * been asking for (non-existent) directory. -ENOENT for you.
3633 if (unlikely(!is_dir && last.name[last.len])) {
3637 if (unlikely(err2)) {
3645 dentry = ERR_PTR(error);
3647 inode_unlock(path->dentry->d_inode);
3649 mnt_drop_write(path->mnt);
3656 struct dentry *kern_path_create(int dfd, const char *pathname,
3657 struct path *path, unsigned int lookup_flags)
3659 return filename_create(dfd, getname_kernel(pathname),
3660 path, lookup_flags);
3662 EXPORT_SYMBOL(kern_path_create);
3664 void done_path_create(struct path *path, struct dentry *dentry)
3667 inode_unlock(path->dentry->d_inode);
3668 mnt_drop_write(path->mnt);
3671 EXPORT_SYMBOL(done_path_create);
3673 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3674 struct path *path, unsigned int lookup_flags)
3676 return filename_create(dfd, getname(pathname), path, lookup_flags);
3678 EXPORT_SYMBOL(user_path_create);
3680 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3682 int error = may_create(dir, dentry);
3687 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3690 if (!dir->i_op->mknod)
3693 error = devcgroup_inode_mknod(mode, dev);
3697 error = security_inode_mknod(dir, dentry, mode, dev);
3701 error = dir->i_op->mknod(dir, dentry, mode, dev);
3703 fsnotify_create(dir, dentry);
3706 EXPORT_SYMBOL(vfs_mknod);
3708 static int may_mknod(umode_t mode)
3710 switch (mode & S_IFMT) {
3716 case 0: /* zero mode translates to S_IFREG */
3725 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3728 struct dentry *dentry;
3731 unsigned int lookup_flags = 0;
3733 error = may_mknod(mode);
3737 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3739 return PTR_ERR(dentry);
3741 if (!IS_POSIXACL(path.dentry->d_inode))
3742 mode &= ~current_umask();
3743 error = security_path_mknod(&path, dentry, mode, dev);
3746 switch (mode & S_IFMT) {
3747 case 0: case S_IFREG:
3748 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3750 ima_post_path_mknod(dentry);
3752 case S_IFCHR: case S_IFBLK:
3753 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3754 new_decode_dev(dev));
3756 case S_IFIFO: case S_IFSOCK:
3757 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3761 done_path_create(&path, dentry);
3762 if (retry_estale(error, lookup_flags)) {
3763 lookup_flags |= LOOKUP_REVAL;
3769 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3772 return do_mknodat(dfd, filename, mode, dev);
3775 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3777 return do_mknodat(AT_FDCWD, filename, mode, dev);
3780 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3782 int error = may_create(dir, dentry);
3783 unsigned max_links = dir->i_sb->s_max_links;
3788 if (!dir->i_op->mkdir)
3791 mode &= (S_IRWXUGO|S_ISVTX);
3792 error = security_inode_mkdir(dir, dentry, mode);
3796 if (max_links && dir->i_nlink >= max_links)
3799 error = dir->i_op->mkdir(dir, dentry, mode);
3801 fsnotify_mkdir(dir, dentry);
3804 EXPORT_SYMBOL(vfs_mkdir);
3806 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3808 struct dentry *dentry;
3811 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3814 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3816 return PTR_ERR(dentry);
3818 if (!IS_POSIXACL(path.dentry->d_inode))
3819 mode &= ~current_umask();
3820 error = security_path_mkdir(&path, dentry, mode);
3822 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3823 done_path_create(&path, dentry);
3824 if (retry_estale(error, lookup_flags)) {
3825 lookup_flags |= LOOKUP_REVAL;
3831 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3833 return do_mkdirat(dfd, pathname, mode);
3836 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3838 return do_mkdirat(AT_FDCWD, pathname, mode);
3841 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3843 int error = may_delete(dir, dentry, 1);
3848 if (!dir->i_op->rmdir)
3852 inode_lock(dentry->d_inode);
3855 if (is_local_mountpoint(dentry))
3858 error = security_inode_rmdir(dir, dentry);
3862 shrink_dcache_parent(dentry);
3863 error = dir->i_op->rmdir(dir, dentry);
3867 dentry->d_inode->i_flags |= S_DEAD;
3869 detach_mounts(dentry);
3872 inode_unlock(dentry->d_inode);
3878 EXPORT_SYMBOL(vfs_rmdir);
3880 long do_rmdir(int dfd, const char __user *pathname)
3883 struct filename *name;
3884 struct dentry *dentry;
3888 unsigned int lookup_flags = 0;
3890 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3891 &path, &last, &type);
3893 return PTR_ERR(name);
3907 error = mnt_want_write(path.mnt);
3911 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3912 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3913 error = PTR_ERR(dentry);
3916 if (!dentry->d_inode) {
3920 error = security_path_rmdir(&path, dentry);
3923 error = vfs_rmdir(path.dentry->d_inode, dentry);
3927 inode_unlock(path.dentry->d_inode);
3928 mnt_drop_write(path.mnt);
3932 if (retry_estale(error, lookup_flags)) {
3933 lookup_flags |= LOOKUP_REVAL;
3939 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3941 return do_rmdir(AT_FDCWD, pathname);
3945 * vfs_unlink - unlink a filesystem object
3946 * @dir: parent directory
3948 * @delegated_inode: returns victim inode, if the inode is delegated.
3950 * The caller must hold dir->i_mutex.
3952 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3953 * return a reference to the inode in delegated_inode. The caller
3954 * should then break the delegation on that inode and retry. Because
3955 * breaking a delegation may take a long time, the caller should drop
3956 * dir->i_mutex before doing so.
3958 * Alternatively, a caller may pass NULL for delegated_inode. This may
3959 * be appropriate for callers that expect the underlying filesystem not
3960 * to be NFS exported.
3962 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3964 struct inode *target = dentry->d_inode;
3965 int error = may_delete(dir, dentry, 0);
3970 if (!dir->i_op->unlink)
3974 if (is_local_mountpoint(dentry))
3977 error = security_inode_unlink(dir, dentry);
3979 error = try_break_deleg(target, delegated_inode);
3982 error = dir->i_op->unlink(dir, dentry);
3985 detach_mounts(dentry);
3990 inode_unlock(target);
3992 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3993 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3994 fsnotify_link_count(target);
4000 EXPORT_SYMBOL(vfs_unlink);
4003 * Make sure that the actual truncation of the file will occur outside its
4004 * directory's i_mutex. Truncate can take a long time if there is a lot of
4005 * writeout happening, and we don't want to prevent access to the directory
4006 * while waiting on the I/O.
4008 long do_unlinkat(int dfd, struct filename *name)
4011 struct dentry *dentry;
4015 struct inode *inode = NULL;
4016 struct inode *delegated_inode = NULL;
4017 unsigned int lookup_flags = 0;
4019 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4021 return PTR_ERR(name);
4024 if (type != LAST_NORM)
4027 error = mnt_want_write(path.mnt);
4031 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4032 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4033 error = PTR_ERR(dentry);
4034 if (!IS_ERR(dentry)) {
4035 /* Why not before? Because we want correct error value */
4036 if (last.name[last.len])
4038 inode = dentry->d_inode;
4039 if (d_is_negative(dentry))
4042 error = security_path_unlink(&path, dentry);
4045 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4049 inode_unlock(path.dentry->d_inode);
4051 iput(inode); /* truncate the inode here */
4053 if (delegated_inode) {
4054 error = break_deleg_wait(&delegated_inode);
4058 mnt_drop_write(path.mnt);
4061 if (retry_estale(error, lookup_flags)) {
4062 lookup_flags |= LOOKUP_REVAL;
4070 if (d_is_negative(dentry))
4072 else if (d_is_dir(dentry))
4079 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4081 if ((flag & ~AT_REMOVEDIR) != 0)
4084 if (flag & AT_REMOVEDIR)
4085 return do_rmdir(dfd, pathname);
4087 return do_unlinkat(dfd, getname(pathname));
4090 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4092 return do_unlinkat(AT_FDCWD, getname(pathname));
4095 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4097 int error = may_create(dir, dentry);
4102 if (!dir->i_op->symlink)
4105 error = security_inode_symlink(dir, dentry, oldname);
4109 error = dir->i_op->symlink(dir, dentry, oldname);
4111 fsnotify_create(dir, dentry);
4114 EXPORT_SYMBOL(vfs_symlink);
4116 long do_symlinkat(const char __user *oldname, int newdfd,
4117 const char __user *newname)
4120 struct filename *from;
4121 struct dentry *dentry;
4123 unsigned int lookup_flags = 0;
4125 from = getname(oldname);
4127 return PTR_ERR(from);
4129 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4130 error = PTR_ERR(dentry);
4134 error = security_path_symlink(&path, dentry, from->name);
4136 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4137 done_path_create(&path, dentry);
4138 if (retry_estale(error, lookup_flags)) {
4139 lookup_flags |= LOOKUP_REVAL;
4147 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4148 int, newdfd, const char __user *, newname)
4150 return do_symlinkat(oldname, newdfd, newname);
4153 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4155 return do_symlinkat(oldname, AT_FDCWD, newname);
4159 * vfs_link - create a new link
4160 * @old_dentry: object to be linked
4162 * @new_dentry: where to create the new link
4163 * @delegated_inode: returns inode needing a delegation break
4165 * The caller must hold dir->i_mutex
4167 * If vfs_link discovers a delegation on the to-be-linked file in need
4168 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4169 * inode in delegated_inode. The caller should then break the delegation
4170 * and retry. Because breaking a delegation may take a long time, the
4171 * caller should drop the i_mutex before doing so.
4173 * Alternatively, a caller may pass NULL for delegated_inode. This may
4174 * be appropriate for callers that expect the underlying filesystem not
4175 * to be NFS exported.
4177 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4179 struct inode *inode = old_dentry->d_inode;
4180 unsigned max_links = dir->i_sb->s_max_links;
4186 error = may_create(dir, new_dentry);
4190 if (dir->i_sb != inode->i_sb)
4194 * A link to an append-only or immutable file cannot be created.
4196 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4199 * Updating the link count will likely cause i_uid and i_gid to
4200 * be writen back improperly if their true value is unknown to
4203 if (HAS_UNMAPPED_ID(inode))
4205 if (!dir->i_op->link)
4207 if (S_ISDIR(inode->i_mode))
4210 error = security_inode_link(old_dentry, dir, new_dentry);
4215 /* Make sure we don't allow creating hardlink to an unlinked file */
4216 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4218 else if (max_links && inode->i_nlink >= max_links)
4221 error = try_break_deleg(inode, delegated_inode);
4223 error = dir->i_op->link(old_dentry, dir, new_dentry);
4226 if (!error && (inode->i_state & I_LINKABLE)) {
4227 spin_lock(&inode->i_lock);
4228 inode->i_state &= ~I_LINKABLE;
4229 spin_unlock(&inode->i_lock);
4231 inode_unlock(inode);
4233 fsnotify_link(dir, inode, new_dentry);
4236 EXPORT_SYMBOL(vfs_link);
4239 * Hardlinks are often used in delicate situations. We avoid
4240 * security-related surprises by not following symlinks on the
4243 * We don't follow them on the oldname either to be compatible
4244 * with linux 2.0, and to avoid hard-linking to directories
4245 * and other special files. --ADM
4247 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4248 const char __user *newname, int flags)
4250 struct dentry *new_dentry;
4251 struct path old_path, new_path;
4252 struct inode *delegated_inode = NULL;
4256 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4259 * To use null names we require CAP_DAC_READ_SEARCH
4260 * This ensures that not everyone will be able to create
4261 * handlink using the passed filedescriptor.
4263 if (flags & AT_EMPTY_PATH) {
4264 if (!capable(CAP_DAC_READ_SEARCH))
4269 if (flags & AT_SYMLINK_FOLLOW)
4270 how |= LOOKUP_FOLLOW;
4272 error = user_path_at(olddfd, oldname, how, &old_path);
4276 new_dentry = user_path_create(newdfd, newname, &new_path,
4277 (how & LOOKUP_REVAL));
4278 error = PTR_ERR(new_dentry);
4279 if (IS_ERR(new_dentry))
4283 if (old_path.mnt != new_path.mnt)
4285 error = may_linkat(&old_path);
4286 if (unlikely(error))
4288 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4291 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4293 done_path_create(&new_path, new_dentry);
4294 if (delegated_inode) {
4295 error = break_deleg_wait(&delegated_inode);
4297 path_put(&old_path);
4301 if (retry_estale(error, how)) {
4302 path_put(&old_path);
4303 how |= LOOKUP_REVAL;
4307 path_put(&old_path);
4312 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4313 int, newdfd, const char __user *, newname, int, flags)
4315 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4318 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4320 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4324 * vfs_rename - rename a filesystem object
4325 * @old_dir: parent of source
4326 * @old_dentry: source
4327 * @new_dir: parent of destination
4328 * @new_dentry: destination
4329 * @delegated_inode: returns an inode needing a delegation break
4330 * @flags: rename flags
4332 * The caller must hold multiple mutexes--see lock_rename()).
4334 * If vfs_rename discovers a delegation in need of breaking at either
4335 * the source or destination, it will return -EWOULDBLOCK and return a
4336 * reference to the inode in delegated_inode. The caller should then
4337 * break the delegation and retry. Because breaking a delegation may
4338 * take a long time, the caller should drop all locks before doing
4341 * Alternatively, a caller may pass NULL for delegated_inode. This may
4342 * be appropriate for callers that expect the underlying filesystem not
4343 * to be NFS exported.
4345 * The worst of all namespace operations - renaming directory. "Perverted"
4346 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4349 * a) we can get into loop creation.
4350 * b) race potential - two innocent renames can create a loop together.
4351 * That's where 4.4 screws up. Current fix: serialization on
4352 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4354 * c) we have to lock _four_ objects - parents and victim (if it exists),
4355 * and source (if it is not a directory).
4356 * And that - after we got ->i_mutex on parents (until then we don't know
4357 * whether the target exists). Solution: try to be smart with locking
4358 * order for inodes. We rely on the fact that tree topology may change
4359 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4360 * move will be locked. Thus we can rank directories by the tree
4361 * (ancestors first) and rank all non-directories after them.
4362 * That works since everybody except rename does "lock parent, lookup,
4363 * lock child" and rename is under ->s_vfs_rename_mutex.
4364 * HOWEVER, it relies on the assumption that any object with ->lookup()
4365 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4366 * we'd better make sure that there's no link(2) for them.
4367 * d) conversion from fhandle to dentry may come in the wrong moment - when
4368 * we are removing the target. Solution: we will have to grab ->i_mutex
4369 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4370 * ->i_mutex on parents, which works but leads to some truly excessive
4373 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4374 struct inode *new_dir, struct dentry *new_dentry,
4375 struct inode **delegated_inode, unsigned int flags)
4378 bool is_dir = d_is_dir(old_dentry);
4379 struct inode *source = old_dentry->d_inode;
4380 struct inode *target = new_dentry->d_inode;
4381 bool new_is_dir = false;
4382 unsigned max_links = new_dir->i_sb->s_max_links;
4383 struct name_snapshot old_name;
4385 if (source == target)
4388 error = may_delete(old_dir, old_dentry, is_dir);
4393 error = may_create(new_dir, new_dentry);
4395 new_is_dir = d_is_dir(new_dentry);
4397 if (!(flags & RENAME_EXCHANGE))
4398 error = may_delete(new_dir, new_dentry, is_dir);
4400 error = may_delete(new_dir, new_dentry, new_is_dir);
4405 if (!old_dir->i_op->rename)
4409 * If we are going to change the parent - check write permissions,
4410 * we'll need to flip '..'.
4412 if (new_dir != old_dir) {
4414 error = inode_permission(source, MAY_WRITE);
4418 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4419 error = inode_permission(target, MAY_WRITE);
4425 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4430 take_dentry_name_snapshot(&old_name, old_dentry);
4432 if (!is_dir || (flags & RENAME_EXCHANGE))
4433 lock_two_nondirectories(source, target);
4438 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4441 if (max_links && new_dir != old_dir) {
4443 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4445 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4446 old_dir->i_nlink >= max_links)
4449 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4450 shrink_dcache_parent(new_dentry);
4452 error = try_break_deleg(source, delegated_inode);
4456 if (target && !new_is_dir) {
4457 error = try_break_deleg(target, delegated_inode);
4461 error = old_dir->i_op->rename(old_dir, old_dentry,
4462 new_dir, new_dentry, flags);
4466 if (!(flags & RENAME_EXCHANGE) && target) {
4468 target->i_flags |= S_DEAD;
4469 dont_mount(new_dentry);
4470 detach_mounts(new_dentry);
4472 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4473 if (!(flags & RENAME_EXCHANGE))
4474 d_move(old_dentry, new_dentry);
4476 d_exchange(old_dentry, new_dentry);
4479 if (!is_dir || (flags & RENAME_EXCHANGE))
4480 unlock_two_nondirectories(source, target);
4482 inode_unlock(target);
4485 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4486 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4487 if (flags & RENAME_EXCHANGE) {
4488 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4489 new_is_dir, NULL, new_dentry);
4492 release_dentry_name_snapshot(&old_name);
4496 EXPORT_SYMBOL(vfs_rename);
4498 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4499 const char __user *newname, unsigned int flags)
4501 struct dentry *old_dentry, *new_dentry;
4502 struct dentry *trap;
4503 struct path old_path, new_path;
4504 struct qstr old_last, new_last;
4505 int old_type, new_type;
4506 struct inode *delegated_inode = NULL;
4507 struct filename *from;
4508 struct filename *to;
4509 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4510 bool should_retry = false;
4513 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4516 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4517 (flags & RENAME_EXCHANGE))
4520 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4523 if (flags & RENAME_EXCHANGE)
4527 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4528 &old_path, &old_last, &old_type);
4530 error = PTR_ERR(from);
4534 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4535 &new_path, &new_last, &new_type);
4537 error = PTR_ERR(to);
4542 if (old_path.mnt != new_path.mnt)
4546 if (old_type != LAST_NORM)
4549 if (flags & RENAME_NOREPLACE)
4551 if (new_type != LAST_NORM)
4554 error = mnt_want_write(old_path.mnt);
4559 trap = lock_rename(new_path.dentry, old_path.dentry);
4561 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4562 error = PTR_ERR(old_dentry);
4563 if (IS_ERR(old_dentry))
4565 /* source must exist */
4567 if (d_is_negative(old_dentry))
4569 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4570 error = PTR_ERR(new_dentry);
4571 if (IS_ERR(new_dentry))
4574 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4576 if (flags & RENAME_EXCHANGE) {
4578 if (d_is_negative(new_dentry))
4581 if (!d_is_dir(new_dentry)) {
4583 if (new_last.name[new_last.len])
4587 /* unless the source is a directory trailing slashes give -ENOTDIR */
4588 if (!d_is_dir(old_dentry)) {
4590 if (old_last.name[old_last.len])
4592 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4595 /* source should not be ancestor of target */
4597 if (old_dentry == trap)
4599 /* target should not be an ancestor of source */
4600 if (!(flags & RENAME_EXCHANGE))
4602 if (new_dentry == trap)
4605 error = security_path_rename(&old_path, old_dentry,
4606 &new_path, new_dentry, flags);
4609 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4610 new_path.dentry->d_inode, new_dentry,
4611 &delegated_inode, flags);
4617 unlock_rename(new_path.dentry, old_path.dentry);
4618 if (delegated_inode) {
4619 error = break_deleg_wait(&delegated_inode);
4623 mnt_drop_write(old_path.mnt);
4625 if (retry_estale(error, lookup_flags))
4626 should_retry = true;
4627 path_put(&new_path);
4630 path_put(&old_path);
4633 should_retry = false;
4634 lookup_flags |= LOOKUP_REVAL;
4641 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4642 int, newdfd, const char __user *, newname, unsigned int, flags)
4644 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4647 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4648 int, newdfd, const char __user *, newname)
4650 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4653 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4655 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4658 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4660 int error = may_create(dir, dentry);
4664 if (!dir->i_op->mknod)
4667 return dir->i_op->mknod(dir, dentry,
4668 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4670 EXPORT_SYMBOL(vfs_whiteout);
4672 int readlink_copy(char __user *buffer, int buflen, const char *link)
4674 int len = PTR_ERR(link);
4679 if (len > (unsigned) buflen)
4681 if (copy_to_user(buffer, link, len))
4688 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4689 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4690 * for any given inode is up to filesystem.
4692 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4695 DEFINE_DELAYED_CALL(done);
4696 struct inode *inode = d_inode(dentry);
4697 const char *link = inode->i_link;
4701 link = inode->i_op->get_link(dentry, inode, &done);
4703 return PTR_ERR(link);
4705 res = readlink_copy(buffer, buflen, link);
4706 do_delayed_call(&done);
4711 * vfs_readlink - copy symlink body into userspace buffer
4712 * @dentry: dentry on which to get symbolic link
4713 * @buffer: user memory pointer
4714 * @buflen: size of buffer
4716 * Does not touch atime. That's up to the caller if necessary
4718 * Does not call security hook.
4720 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4722 struct inode *inode = d_inode(dentry);
4724 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4725 if (unlikely(inode->i_op->readlink))
4726 return inode->i_op->readlink(dentry, buffer, buflen);
4728 if (!d_is_symlink(dentry))
4731 spin_lock(&inode->i_lock);
4732 inode->i_opflags |= IOP_DEFAULT_READLINK;
4733 spin_unlock(&inode->i_lock);
4736 return generic_readlink(dentry, buffer, buflen);
4738 EXPORT_SYMBOL(vfs_readlink);
4741 * vfs_get_link - get symlink body
4742 * @dentry: dentry on which to get symbolic link
4743 * @done: caller needs to free returned data with this
4745 * Calls security hook and i_op->get_link() on the supplied inode.
4747 * It does not touch atime. That's up to the caller if necessary.
4749 * Does not work on "special" symlinks like /proc/$$/fd/N
4751 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4753 const char *res = ERR_PTR(-EINVAL);
4754 struct inode *inode = d_inode(dentry);
4756 if (d_is_symlink(dentry)) {
4757 res = ERR_PTR(security_inode_readlink(dentry));
4759 res = inode->i_op->get_link(dentry, inode, done);
4763 EXPORT_SYMBOL(vfs_get_link);
4765 /* get the link contents into pagecache */
4766 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4767 struct delayed_call *callback)
4771 struct address_space *mapping = inode->i_mapping;
4774 page = find_get_page(mapping, 0);
4776 return ERR_PTR(-ECHILD);
4777 if (!PageUptodate(page)) {
4779 return ERR_PTR(-ECHILD);
4782 page = read_mapping_page(mapping, 0, NULL);
4786 set_delayed_call(callback, page_put_link, page);
4787 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4788 kaddr = page_address(page);
4789 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4793 EXPORT_SYMBOL(page_get_link);
4795 void page_put_link(void *arg)
4799 EXPORT_SYMBOL(page_put_link);
4801 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4803 DEFINE_DELAYED_CALL(done);
4804 int res = readlink_copy(buffer, buflen,
4805 page_get_link(dentry, d_inode(dentry),
4807 do_delayed_call(&done);
4810 EXPORT_SYMBOL(page_readlink);
4813 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4815 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4817 struct address_space *mapping = inode->i_mapping;
4821 unsigned int flags = 0;
4823 flags |= AOP_FLAG_NOFS;
4826 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4827 flags, &page, &fsdata);
4831 memcpy(page_address(page), symname, len-1);
4833 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4840 mark_inode_dirty(inode);
4845 EXPORT_SYMBOL(__page_symlink);
4847 int page_symlink(struct inode *inode, const char *symname, int len)
4849 return __page_symlink(inode, symname, len,
4850 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4852 EXPORT_SYMBOL(page_symlink);
4854 const struct inode_operations page_symlink_inode_operations = {
4855 .get_link = page_get_link,
4857 EXPORT_SYMBOL(page_symlink_inode_operations);