4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
495 #define EMBEDDED_LEVELS 2
500 struct inode *inode; /* path.dentry.d_inode */
505 int total_link_count;
512 } *stack, internal[EMBEDDED_LEVELS];
513 struct filename *name;
514 struct nameidata *saved;
519 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
521 struct nameidata *old = current->nameidata;
522 p->stack = p->internal;
525 p->total_link_count = old ? old->total_link_count : 0;
527 current->nameidata = p;
530 static void restore_nameidata(void)
532 struct nameidata *now = current->nameidata, *old = now->saved;
534 current->nameidata = old;
536 old->total_link_count = now->total_link_count;
537 if (now->stack != now->internal) {
539 now->stack = now->internal;
543 static int __nd_alloc_stack(struct nameidata *nd)
547 if (nd->flags & LOOKUP_RCU) {
548 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
553 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
558 memcpy(p, nd->internal, sizeof(nd->internal));
564 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
565 * @path: nameidate to verify
567 * Rename can sometimes move a file or directory outside of a bind
568 * mount, path_connected allows those cases to be detected.
570 static bool path_connected(const struct path *path)
572 struct vfsmount *mnt = path->mnt;
574 /* Only bind mounts can have disconnected paths */
575 if (mnt->mnt_root == mnt->mnt_sb->s_root)
578 return is_subdir(path->dentry, mnt->mnt_root);
581 static inline int nd_alloc_stack(struct nameidata *nd)
583 if (likely(nd->depth != EMBEDDED_LEVELS))
585 if (likely(nd->stack != nd->internal))
587 return __nd_alloc_stack(nd);
590 static void drop_links(struct nameidata *nd)
594 struct saved *last = nd->stack + i;
595 struct inode *inode = last->inode;
596 if (last->cookie && inode->i_op->put_link) {
597 inode->i_op->put_link(inode, last->cookie);
603 static void terminate_walk(struct nameidata *nd)
606 if (!(nd->flags & LOOKUP_RCU)) {
609 for (i = 0; i < nd->depth; i++)
610 path_put(&nd->stack[i].link);
611 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
616 nd->flags &= ~LOOKUP_RCU;
617 if (!(nd->flags & LOOKUP_ROOT))
624 /* path_put is needed afterwards regardless of success or failure */
625 static bool legitimize_path(struct nameidata *nd,
626 struct path *path, unsigned seq)
628 int res = __legitimize_mnt(path->mnt, nd->m_seq);
635 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
639 return !read_seqcount_retry(&path->dentry->d_seq, seq);
642 static bool legitimize_links(struct nameidata *nd)
645 for (i = 0; i < nd->depth; i++) {
646 struct saved *last = nd->stack + i;
647 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
657 * Path walking has 2 modes, rcu-walk and ref-walk (see
658 * Documentation/filesystems/path-lookup.txt). In situations when we can't
659 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
660 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
661 * mode. Refcounts are grabbed at the last known good point before rcu-walk
662 * got stuck, so ref-walk may continue from there. If this is not successful
663 * (eg. a seqcount has changed), then failure is returned and it's up to caller
664 * to restart the path walk from the beginning in ref-walk mode.
668 * unlazy_walk - try to switch to ref-walk mode.
669 * @nd: nameidata pathwalk data
670 * @dentry: child of nd->path.dentry or NULL
671 * @seq: seq number to check dentry against
672 * Returns: 0 on success, -ECHILD on failure
674 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
675 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
676 * @nd or NULL. Must be called from rcu-walk context.
677 * Nothing should touch nameidata between unlazy_walk() failure and
680 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
682 struct dentry *parent = nd->path.dentry;
684 BUG_ON(!(nd->flags & LOOKUP_RCU));
686 nd->flags &= ~LOOKUP_RCU;
687 if (unlikely(!legitimize_links(nd)))
689 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
691 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
695 * For a negative lookup, the lookup sequence point is the parents
696 * sequence point, and it only needs to revalidate the parent dentry.
698 * For a positive lookup, we need to move both the parent and the
699 * dentry from the RCU domain to be properly refcounted. And the
700 * sequence number in the dentry validates *both* dentry counters,
701 * since we checked the sequence number of the parent after we got
702 * the child sequence number. So we know the parent must still
703 * be valid if the child sequence number is still valid.
706 if (read_seqcount_retry(&parent->d_seq, nd->seq))
708 BUG_ON(nd->inode != parent->d_inode);
710 if (!lockref_get_not_dead(&dentry->d_lockref))
712 if (read_seqcount_retry(&dentry->d_seq, seq))
717 * Sequence counts matched. Now make sure that the root is
718 * still valid and get it if required.
720 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
721 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
738 nd->path.dentry = NULL;
742 if (!(nd->flags & LOOKUP_ROOT))
747 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
749 if (unlikely(!legitimize_path(nd, link, seq))) {
752 nd->flags &= ~LOOKUP_RCU;
754 nd->path.dentry = NULL;
755 if (!(nd->flags & LOOKUP_ROOT))
758 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
765 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
767 return dentry->d_op->d_revalidate(dentry, flags);
771 * complete_walk - successful completion of path walk
772 * @nd: pointer nameidata
774 * If we had been in RCU mode, drop out of it and legitimize nd->path.
775 * Revalidate the final result, unless we'd already done that during
776 * the path walk or the filesystem doesn't ask for it. Return 0 on
777 * success, -error on failure. In case of failure caller does not
778 * need to drop nd->path.
780 static int complete_walk(struct nameidata *nd)
782 struct dentry *dentry = nd->path.dentry;
785 if (nd->flags & LOOKUP_RCU) {
786 if (!(nd->flags & LOOKUP_ROOT))
788 if (unlikely(unlazy_walk(nd, NULL, 0)))
792 if (likely(!(nd->flags & LOOKUP_JUMPED)))
795 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
798 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
808 static void set_root(struct nameidata *nd)
810 get_fs_root(current->fs, &nd->root);
813 static void set_root_rcu(struct nameidata *nd)
815 struct fs_struct *fs = current->fs;
819 seq = read_seqcount_begin(&fs->seq);
821 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
822 } while (read_seqcount_retry(&fs->seq, seq));
825 static void path_put_conditional(struct path *path, struct nameidata *nd)
828 if (path->mnt != nd->path.mnt)
832 static inline void path_to_nameidata(const struct path *path,
833 struct nameidata *nd)
835 if (!(nd->flags & LOOKUP_RCU)) {
836 dput(nd->path.dentry);
837 if (nd->path.mnt != path->mnt)
838 mntput(nd->path.mnt);
840 nd->path.mnt = path->mnt;
841 nd->path.dentry = path->dentry;
845 * Helper to directly jump to a known parsed path from ->follow_link,
846 * caller must have taken a reference to path beforehand.
848 void nd_jump_link(struct path *path)
850 struct nameidata *nd = current->nameidata;
854 nd->inode = nd->path.dentry->d_inode;
855 nd->flags |= LOOKUP_JUMPED;
858 static inline void put_link(struct nameidata *nd)
860 struct saved *last = nd->stack + --nd->depth;
861 struct inode *inode = last->inode;
862 if (last->cookie && inode->i_op->put_link)
863 inode->i_op->put_link(inode, last->cookie);
864 if (!(nd->flags & LOOKUP_RCU))
865 path_put(&last->link);
868 int sysctl_protected_symlinks __read_mostly = 0;
869 int sysctl_protected_hardlinks __read_mostly = 0;
872 * may_follow_link - Check symlink following for unsafe situations
873 * @nd: nameidata pathwalk data
875 * In the case of the sysctl_protected_symlinks sysctl being enabled,
876 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
877 * in a sticky world-writable directory. This is to protect privileged
878 * processes from failing races against path names that may change out
879 * from under them by way of other users creating malicious symlinks.
880 * It will permit symlinks to be followed only when outside a sticky
881 * world-writable directory, or when the uid of the symlink and follower
882 * match, or when the directory owner matches the symlink's owner.
884 * Returns 0 if following the symlink is allowed, -ve on error.
886 static inline int may_follow_link(struct nameidata *nd)
888 const struct inode *inode;
889 const struct inode *parent;
891 if (!sysctl_protected_symlinks)
894 /* Allowed if owner and follower match. */
895 inode = nd->stack[0].inode;
896 if (uid_eq(current_cred()->fsuid, inode->i_uid))
899 /* Allowed if parent directory not sticky and world-writable. */
901 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
904 /* Allowed if parent directory and link owner match. */
905 if (uid_eq(parent->i_uid, inode->i_uid))
908 if (nd->flags & LOOKUP_RCU)
911 audit_log_link_denied("follow_link", &nd->stack[0].link);
916 * safe_hardlink_source - Check for safe hardlink conditions
917 * @inode: the source inode to hardlink from
919 * Return false if at least one of the following conditions:
920 * - inode is not a regular file
922 * - inode is setgid and group-exec
923 * - access failure for read and write
925 * Otherwise returns true.
927 static bool safe_hardlink_source(struct inode *inode)
929 umode_t mode = inode->i_mode;
931 /* Special files should not get pinned to the filesystem. */
935 /* Setuid files should not get pinned to the filesystem. */
939 /* Executable setgid files should not get pinned to the filesystem. */
940 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
943 /* Hardlinking to unreadable or unwritable sources is dangerous. */
944 if (inode_permission(inode, MAY_READ | MAY_WRITE))
951 * may_linkat - Check permissions for creating a hardlink
952 * @link: the source to hardlink from
954 * Block hardlink when all of:
955 * - sysctl_protected_hardlinks enabled
956 * - fsuid does not match inode
957 * - hardlink source is unsafe (see safe_hardlink_source() above)
958 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
960 * Returns 0 if successful, -ve on error.
962 static int may_linkat(struct path *link)
966 if (!sysctl_protected_hardlinks)
969 inode = link->dentry->d_inode;
971 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
972 * otherwise, it must be a safe source.
974 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
977 audit_log_link_denied("linkat", link);
981 static __always_inline
982 const char *get_link(struct nameidata *nd)
984 struct saved *last = nd->stack + nd->depth - 1;
985 struct dentry *dentry = last->link.dentry;
986 struct inode *inode = last->inode;
990 if (!(nd->flags & LOOKUP_RCU)) {
991 touch_atime(&last->link);
993 } else if (atime_needs_update(&last->link, inode)) {
994 if (unlikely(unlazy_walk(nd, NULL, 0)))
995 return ERR_PTR(-ECHILD);
996 touch_atime(&last->link);
999 error = security_inode_follow_link(dentry, inode,
1000 nd->flags & LOOKUP_RCU);
1001 if (unlikely(error))
1002 return ERR_PTR(error);
1004 nd->last_type = LAST_BIND;
1005 res = inode->i_link;
1007 if (nd->flags & LOOKUP_RCU) {
1008 if (unlikely(unlazy_walk(nd, NULL, 0)))
1009 return ERR_PTR(-ECHILD);
1011 res = inode->i_op->follow_link(dentry, &last->cookie);
1012 if (IS_ERR_OR_NULL(res)) {
1013 last->cookie = NULL;
1018 if (nd->flags & LOOKUP_RCU) {
1022 nd->path = nd->root;
1023 d = nd->path.dentry;
1024 nd->inode = d->d_inode;
1025 nd->seq = nd->root_seq;
1026 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1027 return ERR_PTR(-ECHILD);
1031 path_put(&nd->path);
1032 nd->path = nd->root;
1033 path_get(&nd->root);
1034 nd->inode = nd->path.dentry->d_inode;
1036 nd->flags |= LOOKUP_JUMPED;
1037 while (unlikely(*++res == '/'))
1046 * follow_up - Find the mountpoint of path's vfsmount
1048 * Given a path, find the mountpoint of its source file system.
1049 * Replace @path with the path of the mountpoint in the parent mount.
1052 * Return 1 if we went up a level and 0 if we were already at the
1055 int follow_up(struct path *path)
1057 struct mount *mnt = real_mount(path->mnt);
1058 struct mount *parent;
1059 struct dentry *mountpoint;
1061 read_seqlock_excl(&mount_lock);
1062 parent = mnt->mnt_parent;
1063 if (parent == mnt) {
1064 read_sequnlock_excl(&mount_lock);
1067 mntget(&parent->mnt);
1068 mountpoint = dget(mnt->mnt_mountpoint);
1069 read_sequnlock_excl(&mount_lock);
1071 path->dentry = mountpoint;
1073 path->mnt = &parent->mnt;
1076 EXPORT_SYMBOL(follow_up);
1079 * Perform an automount
1080 * - return -EISDIR to tell follow_managed() to stop and return the path we
1083 static int follow_automount(struct path *path, struct nameidata *nd,
1086 struct vfsmount *mnt;
1089 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1092 /* We don't want to mount if someone's just doing a stat -
1093 * unless they're stat'ing a directory and appended a '/' to
1096 * We do, however, want to mount if someone wants to open or
1097 * create a file of any type under the mountpoint, wants to
1098 * traverse through the mountpoint or wants to open the
1099 * mounted directory. Also, autofs may mark negative dentries
1100 * as being automount points. These will need the attentions
1101 * of the daemon to instantiate them before they can be used.
1103 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1104 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1105 path->dentry->d_inode)
1108 nd->total_link_count++;
1109 if (nd->total_link_count >= 40)
1112 mnt = path->dentry->d_op->d_automount(path);
1115 * The filesystem is allowed to return -EISDIR here to indicate
1116 * it doesn't want to automount. For instance, autofs would do
1117 * this so that its userspace daemon can mount on this dentry.
1119 * However, we can only permit this if it's a terminal point in
1120 * the path being looked up; if it wasn't then the remainder of
1121 * the path is inaccessible and we should say so.
1123 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1125 return PTR_ERR(mnt);
1128 if (!mnt) /* mount collision */
1131 if (!*need_mntput) {
1132 /* lock_mount() may release path->mnt on error */
1134 *need_mntput = true;
1136 err = finish_automount(mnt, path);
1140 /* Someone else made a mount here whilst we were busy */
1145 path->dentry = dget(mnt->mnt_root);
1154 * Handle a dentry that is managed in some way.
1155 * - Flagged for transit management (autofs)
1156 * - Flagged as mountpoint
1157 * - Flagged as automount point
1159 * This may only be called in refwalk mode.
1161 * Serialization is taken care of in namespace.c
1163 static int follow_managed(struct path *path, struct nameidata *nd)
1165 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1167 bool need_mntput = false;
1170 /* Given that we're not holding a lock here, we retain the value in a
1171 * local variable for each dentry as we look at it so that we don't see
1172 * the components of that value change under us */
1173 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1174 managed &= DCACHE_MANAGED_DENTRY,
1175 unlikely(managed != 0)) {
1176 /* Allow the filesystem to manage the transit without i_mutex
1178 if (managed & DCACHE_MANAGE_TRANSIT) {
1179 BUG_ON(!path->dentry->d_op);
1180 BUG_ON(!path->dentry->d_op->d_manage);
1181 ret = path->dentry->d_op->d_manage(path->dentry, false);
1186 /* Transit to a mounted filesystem. */
1187 if (managed & DCACHE_MOUNTED) {
1188 struct vfsmount *mounted = lookup_mnt(path);
1193 path->mnt = mounted;
1194 path->dentry = dget(mounted->mnt_root);
1199 /* Something is mounted on this dentry in another
1200 * namespace and/or whatever was mounted there in this
1201 * namespace got unmounted before lookup_mnt() could
1205 /* Handle an automount point */
1206 if (managed & DCACHE_NEED_AUTOMOUNT) {
1207 ret = follow_automount(path, nd, &need_mntput);
1213 /* We didn't change the current path point */
1217 if (need_mntput && path->mnt == mnt)
1222 nd->flags |= LOOKUP_JUMPED;
1223 if (unlikely(ret < 0))
1224 path_put_conditional(path, nd);
1228 int follow_down_one(struct path *path)
1230 struct vfsmount *mounted;
1232 mounted = lookup_mnt(path);
1236 path->mnt = mounted;
1237 path->dentry = dget(mounted->mnt_root);
1242 EXPORT_SYMBOL(follow_down_one);
1244 static inline int managed_dentry_rcu(struct dentry *dentry)
1246 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1247 dentry->d_op->d_manage(dentry, true) : 0;
1251 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1252 * we meet a managed dentry that would need blocking.
1254 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1255 struct inode **inode, unsigned *seqp)
1258 struct mount *mounted;
1260 * Don't forget we might have a non-mountpoint managed dentry
1261 * that wants to block transit.
1263 switch (managed_dentry_rcu(path->dentry)) {
1273 if (!d_mountpoint(path->dentry))
1274 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1276 mounted = __lookup_mnt(path->mnt, path->dentry);
1279 path->mnt = &mounted->mnt;
1280 path->dentry = mounted->mnt.mnt_root;
1281 nd->flags |= LOOKUP_JUMPED;
1282 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1284 * Update the inode too. We don't need to re-check the
1285 * dentry sequence number here after this d_inode read,
1286 * because a mount-point is always pinned.
1288 *inode = path->dentry->d_inode;
1290 return !read_seqretry(&mount_lock, nd->m_seq) &&
1291 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1294 static int follow_dotdot_rcu(struct nameidata *nd)
1296 struct inode *inode = nd->inode;
1301 if (path_equal(&nd->path, &nd->root))
1303 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1304 struct dentry *old = nd->path.dentry;
1305 struct dentry *parent = old->d_parent;
1308 inode = parent->d_inode;
1309 seq = read_seqcount_begin(&parent->d_seq);
1310 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1312 nd->path.dentry = parent;
1314 if (unlikely(!path_connected(&nd->path)))
1318 struct mount *mnt = real_mount(nd->path.mnt);
1319 struct mount *mparent = mnt->mnt_parent;
1320 struct dentry *mountpoint = mnt->mnt_mountpoint;
1321 struct inode *inode2 = mountpoint->d_inode;
1322 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1323 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1325 if (&mparent->mnt == nd->path.mnt)
1327 /* we know that mountpoint was pinned */
1328 nd->path.dentry = mountpoint;
1329 nd->path.mnt = &mparent->mnt;
1334 while (unlikely(d_mountpoint(nd->path.dentry))) {
1335 struct mount *mounted;
1336 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1337 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341 nd->path.mnt = &mounted->mnt;
1342 nd->path.dentry = mounted->mnt.mnt_root;
1343 inode = nd->path.dentry->d_inode;
1344 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1351 * Follow down to the covering mount currently visible to userspace. At each
1352 * point, the filesystem owning that dentry may be queried as to whether the
1353 * caller is permitted to proceed or not.
1355 int follow_down(struct path *path)
1360 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1361 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1362 /* Allow the filesystem to manage the transit without i_mutex
1365 * We indicate to the filesystem if someone is trying to mount
1366 * something here. This gives autofs the chance to deny anyone
1367 * other than its daemon the right to mount on its
1370 * The filesystem may sleep at this point.
1372 if (managed & DCACHE_MANAGE_TRANSIT) {
1373 BUG_ON(!path->dentry->d_op);
1374 BUG_ON(!path->dentry->d_op->d_manage);
1375 ret = path->dentry->d_op->d_manage(
1376 path->dentry, false);
1378 return ret == -EISDIR ? 0 : ret;
1381 /* Transit to a mounted filesystem. */
1382 if (managed & DCACHE_MOUNTED) {
1383 struct vfsmount *mounted = lookup_mnt(path);
1388 path->mnt = mounted;
1389 path->dentry = dget(mounted->mnt_root);
1393 /* Don't handle automount points here */
1398 EXPORT_SYMBOL(follow_down);
1401 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1403 static void follow_mount(struct path *path)
1405 while (d_mountpoint(path->dentry)) {
1406 struct vfsmount *mounted = lookup_mnt(path);
1411 path->mnt = mounted;
1412 path->dentry = dget(mounted->mnt_root);
1416 static int follow_dotdot(struct nameidata *nd)
1422 struct dentry *old = nd->path.dentry;
1424 if (nd->path.dentry == nd->root.dentry &&
1425 nd->path.mnt == nd->root.mnt) {
1428 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1429 /* rare case of legitimate dget_parent()... */
1430 nd->path.dentry = dget_parent(nd->path.dentry);
1432 if (unlikely(!path_connected(&nd->path)))
1436 if (!follow_up(&nd->path))
1439 follow_mount(&nd->path);
1440 nd->inode = nd->path.dentry->d_inode;
1445 * This looks up the name in dcache, possibly revalidates the old dentry and
1446 * allocates a new one if not found or not valid. In the need_lookup argument
1447 * returns whether i_op->lookup is necessary.
1449 * dir->d_inode->i_mutex must be held
1451 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1452 unsigned int flags, bool *need_lookup)
1454 struct dentry *dentry;
1457 *need_lookup = false;
1458 dentry = d_lookup(dir, name);
1460 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1461 error = d_revalidate(dentry, flags);
1462 if (unlikely(error <= 0)) {
1465 return ERR_PTR(error);
1467 d_invalidate(dentry);
1476 dentry = d_alloc(dir, name);
1477 if (unlikely(!dentry))
1478 return ERR_PTR(-ENOMEM);
1480 *need_lookup = true;
1486 * Call i_op->lookup on the dentry. The dentry must be negative and
1489 * dir->d_inode->i_mutex must be held
1491 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1496 /* Don't create child dentry for a dead directory. */
1497 if (unlikely(IS_DEADDIR(dir))) {
1499 return ERR_PTR(-ENOENT);
1502 old = dir->i_op->lookup(dir, dentry, flags);
1503 if (unlikely(old)) {
1510 static struct dentry *__lookup_hash(struct qstr *name,
1511 struct dentry *base, unsigned int flags)
1514 struct dentry *dentry;
1516 dentry = lookup_dcache(name, base, flags, &need_lookup);
1520 return lookup_real(base->d_inode, dentry, flags);
1524 * It's more convoluted than I'd like it to be, but... it's still fairly
1525 * small and for now I'd prefer to have fast path as straight as possible.
1526 * It _is_ time-critical.
1528 static int lookup_fast(struct nameidata *nd,
1529 struct path *path, struct inode **inode,
1532 struct vfsmount *mnt = nd->path.mnt;
1533 struct dentry *dentry, *parent = nd->path.dentry;
1539 * Rename seqlock is not required here because in the off chance
1540 * of a false negative due to a concurrent rename, we're going to
1541 * do the non-racy lookup, below.
1543 if (nd->flags & LOOKUP_RCU) {
1546 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1551 * This sequence count validates that the inode matches
1552 * the dentry name information from lookup.
1554 *inode = d_backing_inode(dentry);
1555 negative = d_is_negative(dentry);
1556 if (read_seqcount_retry(&dentry->d_seq, seq))
1560 * This sequence count validates that the parent had no
1561 * changes while we did the lookup of the dentry above.
1563 * The memory barrier in read_seqcount_begin of child is
1564 * enough, we can use __read_seqcount_retry here.
1566 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1570 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1571 status = d_revalidate(dentry, nd->flags);
1572 if (unlikely(status <= 0)) {
1573 if (status != -ECHILD)
1579 * Note: do negative dentry check after revalidation in
1580 * case that drops it.
1585 path->dentry = dentry;
1586 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1589 if (unlazy_walk(nd, dentry, seq))
1592 dentry = __d_lookup(parent, &nd->last);
1595 if (unlikely(!dentry))
1598 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1599 status = d_revalidate(dentry, nd->flags);
1600 if (unlikely(status <= 0)) {
1605 d_invalidate(dentry);
1610 if (unlikely(d_is_negative(dentry))) {
1615 path->dentry = dentry;
1616 err = follow_managed(path, nd);
1618 *inode = d_backing_inode(path->dentry);
1625 /* Fast lookup failed, do it the slow way */
1626 static int lookup_slow(struct nameidata *nd, struct path *path)
1628 struct dentry *dentry, *parent;
1630 parent = nd->path.dentry;
1631 BUG_ON(nd->inode != parent->d_inode);
1633 mutex_lock(&parent->d_inode->i_mutex);
1634 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1635 mutex_unlock(&parent->d_inode->i_mutex);
1637 return PTR_ERR(dentry);
1638 path->mnt = nd->path.mnt;
1639 path->dentry = dentry;
1640 return follow_managed(path, nd);
1643 static inline int may_lookup(struct nameidata *nd)
1645 if (nd->flags & LOOKUP_RCU) {
1646 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1649 if (unlazy_walk(nd, NULL, 0))
1652 return inode_permission(nd->inode, MAY_EXEC);
1655 static inline int handle_dots(struct nameidata *nd, int type)
1657 if (type == LAST_DOTDOT) {
1658 if (nd->flags & LOOKUP_RCU) {
1659 return follow_dotdot_rcu(nd);
1661 return follow_dotdot(nd);
1666 static int pick_link(struct nameidata *nd, struct path *link,
1667 struct inode *inode, unsigned seq)
1671 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1672 path_to_nameidata(link, nd);
1675 if (!(nd->flags & LOOKUP_RCU)) {
1676 if (link->mnt == nd->path.mnt)
1679 error = nd_alloc_stack(nd);
1680 if (unlikely(error)) {
1681 if (error == -ECHILD) {
1682 if (unlikely(unlazy_link(nd, link, seq)))
1684 error = nd_alloc_stack(nd);
1692 last = nd->stack + nd->depth++;
1694 last->cookie = NULL;
1695 last->inode = inode;
1701 * Do we need to follow links? We _really_ want to be able
1702 * to do this check without having to look at inode->i_op,
1703 * so we keep a cache of "no, this doesn't need follow_link"
1704 * for the common case.
1706 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1708 struct inode *inode, unsigned seq)
1710 if (likely(!d_is_symlink(link->dentry)))
1714 /* make sure that d_is_symlink above matches inode */
1715 if (nd->flags & LOOKUP_RCU) {
1716 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1719 return pick_link(nd, link, inode, seq);
1722 enum {WALK_GET = 1, WALK_PUT = 2};
1724 static int walk_component(struct nameidata *nd, int flags)
1727 struct inode *inode;
1731 * "." and ".." are special - ".." especially so because it has
1732 * to be able to know about the current root directory and
1733 * parent relationships.
1735 if (unlikely(nd->last_type != LAST_NORM)) {
1736 err = handle_dots(nd, nd->last_type);
1737 if (flags & WALK_PUT)
1741 err = lookup_fast(nd, &path, &inode, &seq);
1742 if (unlikely(err)) {
1746 err = lookup_slow(nd, &path);
1750 seq = 0; /* we are already out of RCU mode */
1752 if (d_is_negative(path.dentry))
1754 inode = d_backing_inode(path.dentry);
1757 if (flags & WALK_PUT)
1759 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1762 path_to_nameidata(&path, nd);
1768 path_to_nameidata(&path, nd);
1773 * We can do the critical dentry name comparison and hashing
1774 * operations one word at a time, but we are limited to:
1776 * - Architectures with fast unaligned word accesses. We could
1777 * do a "get_unaligned()" if this helps and is sufficiently
1780 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1781 * do not trap on the (extremely unlikely) case of a page
1782 * crossing operation.
1784 * - Furthermore, we need an efficient 64-bit compile for the
1785 * 64-bit case in order to generate the "number of bytes in
1786 * the final mask". Again, that could be replaced with a
1787 * efficient population count instruction or similar.
1789 #ifdef CONFIG_DCACHE_WORD_ACCESS
1791 #include <asm/word-at-a-time.h>
1795 static inline unsigned int fold_hash(unsigned long hash)
1797 return hash_64(hash, 32);
1800 #else /* 32-bit case */
1802 #define fold_hash(x) (x)
1806 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1808 unsigned long a, mask;
1809 unsigned long hash = 0;
1812 a = load_unaligned_zeropad(name);
1813 if (len < sizeof(unsigned long))
1817 name += sizeof(unsigned long);
1818 len -= sizeof(unsigned long);
1822 mask = bytemask_from_count(len);
1825 return fold_hash(hash);
1827 EXPORT_SYMBOL(full_name_hash);
1830 * Calculate the length and hash of the path component, and
1831 * return the "hash_len" as the result.
1833 static inline u64 hash_name(const char *name)
1835 unsigned long a, b, adata, bdata, mask, hash, len;
1836 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1839 len = -sizeof(unsigned long);
1841 hash = (hash + a) * 9;
1842 len += sizeof(unsigned long);
1843 a = load_unaligned_zeropad(name+len);
1844 b = a ^ REPEAT_BYTE('/');
1845 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1847 adata = prep_zero_mask(a, adata, &constants);
1848 bdata = prep_zero_mask(b, bdata, &constants);
1850 mask = create_zero_mask(adata | bdata);
1852 hash += a & zero_bytemask(mask);
1853 len += find_zero(mask);
1854 return hashlen_create(fold_hash(hash), len);
1859 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1861 unsigned long hash = init_name_hash();
1863 hash = partial_name_hash(*name++, hash);
1864 return end_name_hash(hash);
1866 EXPORT_SYMBOL(full_name_hash);
1869 * We know there's a real path component here of at least
1872 static inline u64 hash_name(const char *name)
1874 unsigned long hash = init_name_hash();
1875 unsigned long len = 0, c;
1877 c = (unsigned char)*name;
1880 hash = partial_name_hash(c, hash);
1881 c = (unsigned char)name[len];
1882 } while (c && c != '/');
1883 return hashlen_create(end_name_hash(hash), len);
1890 * This is the basic name resolution function, turning a pathname into
1891 * the final dentry. We expect 'base' to be positive and a directory.
1893 * Returns 0 and nd will have valid dentry and mnt on success.
1894 * Returns error and drops reference to input namei data on failure.
1896 static int link_path_walk(const char *name, struct nameidata *nd)
1905 /* At this point we know we have a real path component. */
1910 err = may_lookup(nd);
1914 hash_len = hash_name(name);
1917 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1919 if (name[1] == '.') {
1921 nd->flags |= LOOKUP_JUMPED;
1927 if (likely(type == LAST_NORM)) {
1928 struct dentry *parent = nd->path.dentry;
1929 nd->flags &= ~LOOKUP_JUMPED;
1930 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1931 struct qstr this = { { .hash_len = hash_len }, .name = name };
1932 err = parent->d_op->d_hash(parent, &this);
1935 hash_len = this.hash_len;
1940 nd->last.hash_len = hash_len;
1941 nd->last.name = name;
1942 nd->last_type = type;
1944 name += hashlen_len(hash_len);
1948 * If it wasn't NUL, we know it was '/'. Skip that
1949 * slash, and continue until no more slashes.
1953 } while (unlikely(*name == '/'));
1954 if (unlikely(!*name)) {
1956 /* pathname body, done */
1959 name = nd->stack[nd->depth - 1].name;
1960 /* trailing symlink, done */
1963 /* last component of nested symlink */
1964 err = walk_component(nd, WALK_GET | WALK_PUT);
1966 err = walk_component(nd, WALK_GET);
1972 const char *s = get_link(nd);
1981 nd->stack[nd->depth - 1].name = name;
1986 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1987 if (nd->flags & LOOKUP_RCU) {
1988 if (unlazy_walk(nd, NULL, 0))
1996 static const char *path_init(struct nameidata *nd, unsigned flags)
1999 const char *s = nd->name->name;
2001 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2002 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2004 if (flags & LOOKUP_ROOT) {
2005 struct dentry *root = nd->root.dentry;
2006 struct inode *inode = root->d_inode;
2008 if (!d_can_lookup(root))
2009 return ERR_PTR(-ENOTDIR);
2010 retval = inode_permission(inode, MAY_EXEC);
2012 return ERR_PTR(retval);
2014 nd->path = nd->root;
2016 if (flags & LOOKUP_RCU) {
2018 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2019 nd->root_seq = nd->seq;
2020 nd->m_seq = read_seqbegin(&mount_lock);
2022 path_get(&nd->path);
2027 nd->root.mnt = NULL;
2029 nd->m_seq = read_seqbegin(&mount_lock);
2031 if (flags & LOOKUP_RCU) {
2034 nd->seq = nd->root_seq;
2037 path_get(&nd->root);
2039 nd->path = nd->root;
2040 } else if (nd->dfd == AT_FDCWD) {
2041 if (flags & LOOKUP_RCU) {
2042 struct fs_struct *fs = current->fs;
2048 seq = read_seqcount_begin(&fs->seq);
2050 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2051 } while (read_seqcount_retry(&fs->seq, seq));
2053 get_fs_pwd(current->fs, &nd->path);
2056 /* Caller must check execute permissions on the starting path component */
2057 struct fd f = fdget_raw(nd->dfd);
2058 struct dentry *dentry;
2061 return ERR_PTR(-EBADF);
2063 dentry = f.file->f_path.dentry;
2066 if (!d_can_lookup(dentry)) {
2068 return ERR_PTR(-ENOTDIR);
2072 nd->path = f.file->f_path;
2073 if (flags & LOOKUP_RCU) {
2075 nd->inode = nd->path.dentry->d_inode;
2076 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2078 path_get(&nd->path);
2079 nd->inode = nd->path.dentry->d_inode;
2085 nd->inode = nd->path.dentry->d_inode;
2086 if (!(flags & LOOKUP_RCU))
2088 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2090 if (!(nd->flags & LOOKUP_ROOT))
2091 nd->root.mnt = NULL;
2093 return ERR_PTR(-ECHILD);
2096 static const char *trailing_symlink(struct nameidata *nd)
2099 int error = may_follow_link(nd);
2100 if (unlikely(error))
2101 return ERR_PTR(error);
2102 nd->flags |= LOOKUP_PARENT;
2103 nd->stack[0].name = NULL;
2108 static inline int lookup_last(struct nameidata *nd)
2110 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2111 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2113 nd->flags &= ~LOOKUP_PARENT;
2114 return walk_component(nd,
2115 nd->flags & LOOKUP_FOLLOW
2117 ? WALK_PUT | WALK_GET
2122 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2123 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2125 const char *s = path_init(nd, flags);
2130 while (!(err = link_path_walk(s, nd))
2131 && ((err = lookup_last(nd)) > 0)) {
2132 s = trailing_symlink(nd);
2139 err = complete_walk(nd);
2141 if (!err && nd->flags & LOOKUP_DIRECTORY)
2142 if (!d_can_lookup(nd->path.dentry))
2146 nd->path.mnt = NULL;
2147 nd->path.dentry = NULL;
2153 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2154 struct path *path, struct path *root)
2157 struct nameidata nd;
2159 return PTR_ERR(name);
2160 if (unlikely(root)) {
2162 flags |= LOOKUP_ROOT;
2164 set_nameidata(&nd, dfd, name);
2165 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2166 if (unlikely(retval == -ECHILD))
2167 retval = path_lookupat(&nd, flags, path);
2168 if (unlikely(retval == -ESTALE))
2169 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2171 if (likely(!retval))
2172 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2173 restore_nameidata();
2178 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2179 static int path_parentat(struct nameidata *nd, unsigned flags,
2180 struct path *parent)
2182 const char *s = path_init(nd, flags);
2186 err = link_path_walk(s, nd);
2188 err = complete_walk(nd);
2191 nd->path.mnt = NULL;
2192 nd->path.dentry = NULL;
2198 static struct filename *filename_parentat(int dfd, struct filename *name,
2199 unsigned int flags, struct path *parent,
2200 struct qstr *last, int *type)
2203 struct nameidata nd;
2207 set_nameidata(&nd, dfd, name);
2208 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2209 if (unlikely(retval == -ECHILD))
2210 retval = path_parentat(&nd, flags, parent);
2211 if (unlikely(retval == -ESTALE))
2212 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2213 if (likely(!retval)) {
2215 *type = nd.last_type;
2216 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2219 name = ERR_PTR(retval);
2221 restore_nameidata();
2225 /* does lookup, returns the object with parent locked */
2226 struct dentry *kern_path_locked(const char *name, struct path *path)
2228 struct filename *filename;
2233 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2235 if (IS_ERR(filename))
2236 return ERR_CAST(filename);
2237 if (unlikely(type != LAST_NORM)) {
2240 return ERR_PTR(-EINVAL);
2242 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2243 d = __lookup_hash(&last, path->dentry, 0);
2245 mutex_unlock(&path->dentry->d_inode->i_mutex);
2252 int kern_path(const char *name, unsigned int flags, struct path *path)
2254 return filename_lookup(AT_FDCWD, getname_kernel(name),
2257 EXPORT_SYMBOL(kern_path);
2260 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2261 * @dentry: pointer to dentry of the base directory
2262 * @mnt: pointer to vfs mount of the base directory
2263 * @name: pointer to file name
2264 * @flags: lookup flags
2265 * @path: pointer to struct path to fill
2267 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2268 const char *name, unsigned int flags,
2271 struct path root = {.mnt = mnt, .dentry = dentry};
2272 /* the first argument of filename_lookup() is ignored with root */
2273 return filename_lookup(AT_FDCWD, getname_kernel(name),
2274 flags , path, &root);
2276 EXPORT_SYMBOL(vfs_path_lookup);
2279 * lookup_one_len - filesystem helper to lookup single pathname component
2280 * @name: pathname component to lookup
2281 * @base: base directory to lookup from
2282 * @len: maximum length @len should be interpreted to
2284 * Note that this routine is purely a helper for filesystem usage and should
2285 * not be called by generic code.
2287 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2293 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2297 this.hash = full_name_hash(name, len);
2299 return ERR_PTR(-EACCES);
2301 if (unlikely(name[0] == '.')) {
2302 if (len < 2 || (len == 2 && name[1] == '.'))
2303 return ERR_PTR(-EACCES);
2307 c = *(const unsigned char *)name++;
2308 if (c == '/' || c == '\0')
2309 return ERR_PTR(-EACCES);
2312 * See if the low-level filesystem might want
2313 * to use its own hash..
2315 if (base->d_flags & DCACHE_OP_HASH) {
2316 int err = base->d_op->d_hash(base, &this);
2318 return ERR_PTR(err);
2321 err = inode_permission(base->d_inode, MAY_EXEC);
2323 return ERR_PTR(err);
2325 return __lookup_hash(&this, base, 0);
2327 EXPORT_SYMBOL(lookup_one_len);
2329 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2330 struct path *path, int *empty)
2332 return filename_lookup(dfd, getname_flags(name, flags, empty),
2335 EXPORT_SYMBOL(user_path_at_empty);
2338 * NB: most callers don't do anything directly with the reference to the
2339 * to struct filename, but the nd->last pointer points into the name string
2340 * allocated by getname. So we must hold the reference to it until all
2341 * path-walking is complete.
2343 static inline struct filename *
2344 user_path_parent(int dfd, const char __user *path,
2345 struct path *parent,
2350 /* only LOOKUP_REVAL is allowed in extra flags */
2351 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2352 parent, last, type);
2356 * mountpoint_last - look up last component for umount
2357 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2358 * @path: pointer to container for result
2360 * This is a special lookup_last function just for umount. In this case, we
2361 * need to resolve the path without doing any revalidation.
2363 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2364 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2365 * in almost all cases, this lookup will be served out of the dcache. The only
2366 * cases where it won't are if nd->last refers to a symlink or the path is
2367 * bogus and it doesn't exist.
2370 * -error: if there was an error during lookup. This includes -ENOENT if the
2371 * lookup found a negative dentry. The nd->path reference will also be
2374 * 0: if we successfully resolved nd->path and found it to not to be a
2375 * symlink that needs to be followed. "path" will also be populated.
2376 * The nd->path reference will also be put.
2378 * 1: if we successfully resolved nd->last and found it to be a symlink
2379 * that needs to be followed. "path" will be populated with the path
2380 * to the link, and nd->path will *not* be put.
2383 mountpoint_last(struct nameidata *nd, struct path *path)
2386 struct dentry *dentry;
2387 struct dentry *dir = nd->path.dentry;
2389 /* If we're in rcuwalk, drop out of it to handle last component */
2390 if (nd->flags & LOOKUP_RCU) {
2391 if (unlazy_walk(nd, NULL, 0))
2395 nd->flags &= ~LOOKUP_PARENT;
2397 if (unlikely(nd->last_type != LAST_NORM)) {
2398 error = handle_dots(nd, nd->last_type);
2401 dentry = dget(nd->path.dentry);
2405 mutex_lock(&dir->d_inode->i_mutex);
2406 dentry = d_lookup(dir, &nd->last);
2409 * No cached dentry. Mounted dentries are pinned in the cache,
2410 * so that means that this dentry is probably a symlink or the
2411 * path doesn't actually point to a mounted dentry.
2413 dentry = d_alloc(dir, &nd->last);
2415 mutex_unlock(&dir->d_inode->i_mutex);
2418 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2419 if (IS_ERR(dentry)) {
2420 mutex_unlock(&dir->d_inode->i_mutex);
2421 return PTR_ERR(dentry);
2424 mutex_unlock(&dir->d_inode->i_mutex);
2427 if (d_is_negative(dentry)) {
2433 path->dentry = dentry;
2434 path->mnt = nd->path.mnt;
2435 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2436 d_backing_inode(dentry), 0);
2437 if (unlikely(error))
2445 * path_mountpoint - look up a path to be umounted
2446 * @nd: lookup context
2447 * @flags: lookup flags
2448 * @path: pointer to container for result
2450 * Look up the given name, but don't attempt to revalidate the last component.
2451 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2454 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2456 const char *s = path_init(nd, flags);
2460 while (!(err = link_path_walk(s, nd)) &&
2461 (err = mountpoint_last(nd, path)) > 0) {
2462 s = trailing_symlink(nd);
2473 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2476 struct nameidata nd;
2479 return PTR_ERR(name);
2480 set_nameidata(&nd, dfd, name);
2481 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2482 if (unlikely(error == -ECHILD))
2483 error = path_mountpoint(&nd, flags, path);
2484 if (unlikely(error == -ESTALE))
2485 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2487 audit_inode(name, path->dentry, 0);
2488 restore_nameidata();
2494 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2495 * @dfd: directory file descriptor
2496 * @name: pathname from userland
2497 * @flags: lookup flags
2498 * @path: pointer to container to hold result
2500 * A umount is a special case for path walking. We're not actually interested
2501 * in the inode in this situation, and ESTALE errors can be a problem. We
2502 * simply want track down the dentry and vfsmount attached at the mountpoint
2503 * and avoid revalidating the last component.
2505 * Returns 0 and populates "path" on success.
2508 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2511 return filename_mountpoint(dfd, getname(name), path, flags);
2515 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2518 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2520 EXPORT_SYMBOL(kern_path_mountpoint);
2522 int __check_sticky(struct inode *dir, struct inode *inode)
2524 kuid_t fsuid = current_fsuid();
2526 if (uid_eq(inode->i_uid, fsuid))
2528 if (uid_eq(dir->i_uid, fsuid))
2530 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2532 EXPORT_SYMBOL(__check_sticky);
2535 * Check whether we can remove a link victim from directory dir, check
2536 * whether the type of victim is right.
2537 * 1. We can't do it if dir is read-only (done in permission())
2538 * 2. We should have write and exec permissions on dir
2539 * 3. We can't remove anything from append-only dir
2540 * 4. We can't do anything with immutable dir (done in permission())
2541 * 5. If the sticky bit on dir is set we should either
2542 * a. be owner of dir, or
2543 * b. be owner of victim, or
2544 * c. have CAP_FOWNER capability
2545 * 6. If the victim is append-only or immutable we can't do antyhing with
2546 * links pointing to it.
2547 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2548 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2549 * 9. We can't remove a root or mountpoint.
2550 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2551 * nfs_async_unlink().
2553 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2555 struct inode *inode = d_backing_inode(victim);
2558 if (d_is_negative(victim))
2562 BUG_ON(victim->d_parent->d_inode != dir);
2563 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2565 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2571 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2572 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2575 if (!d_is_dir(victim))
2577 if (IS_ROOT(victim))
2579 } else if (d_is_dir(victim))
2581 if (IS_DEADDIR(dir))
2583 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2588 /* Check whether we can create an object with dentry child in directory
2590 * 1. We can't do it if child already exists (open has special treatment for
2591 * this case, but since we are inlined it's OK)
2592 * 2. We can't do it if dir is read-only (done in permission())
2593 * 3. We should have write and exec permissions on dir
2594 * 4. We can't do it if dir is immutable (done in permission())
2596 static inline int may_create(struct inode *dir, struct dentry *child)
2598 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2601 if (IS_DEADDIR(dir))
2603 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2607 * p1 and p2 should be directories on the same fs.
2609 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2614 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2618 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2620 p = d_ancestor(p2, p1);
2622 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2623 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2627 p = d_ancestor(p1, p2);
2629 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2630 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2634 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2635 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2638 EXPORT_SYMBOL(lock_rename);
2640 void unlock_rename(struct dentry *p1, struct dentry *p2)
2642 mutex_unlock(&p1->d_inode->i_mutex);
2644 mutex_unlock(&p2->d_inode->i_mutex);
2645 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2648 EXPORT_SYMBOL(unlock_rename);
2650 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2653 int error = may_create(dir, dentry);
2657 if (!dir->i_op->create)
2658 return -EACCES; /* shouldn't it be ENOSYS? */
2661 error = security_inode_create(dir, dentry, mode);
2664 error = dir->i_op->create(dir, dentry, mode, want_excl);
2666 fsnotify_create(dir, dentry);
2669 EXPORT_SYMBOL(vfs_create);
2671 static int may_open(struct path *path, int acc_mode, int flag)
2673 struct dentry *dentry = path->dentry;
2674 struct inode *inode = dentry->d_inode;
2684 switch (inode->i_mode & S_IFMT) {
2688 if (acc_mode & MAY_WRITE)
2693 if (path->mnt->mnt_flags & MNT_NODEV)
2702 error = inode_permission(inode, acc_mode);
2707 * An append-only file must be opened in append mode for writing.
2709 if (IS_APPEND(inode)) {
2710 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2716 /* O_NOATIME can only be set by the owner or superuser */
2717 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2723 static int handle_truncate(struct file *filp)
2725 struct path *path = &filp->f_path;
2726 struct inode *inode = path->dentry->d_inode;
2727 int error = get_write_access(inode);
2731 * Refuse to truncate files with mandatory locks held on them.
2733 error = locks_verify_locked(filp);
2735 error = security_path_truncate(path);
2737 error = do_truncate(path->dentry, 0,
2738 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2741 put_write_access(inode);
2745 static inline int open_to_namei_flags(int flag)
2747 if ((flag & O_ACCMODE) == 3)
2752 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2754 int error = security_path_mknod(dir, dentry, mode, 0);
2758 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2762 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2766 * Attempt to atomically look up, create and open a file from a negative
2769 * Returns 0 if successful. The file will have been created and attached to
2770 * @file by the filesystem calling finish_open().
2772 * Returns 1 if the file was looked up only or didn't need creating. The
2773 * caller will need to perform the open themselves. @path will have been
2774 * updated to point to the new dentry. This may be negative.
2776 * Returns an error code otherwise.
2778 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2779 struct path *path, struct file *file,
2780 const struct open_flags *op,
2781 bool got_write, bool need_lookup,
2784 struct inode *dir = nd->path.dentry->d_inode;
2785 unsigned open_flag = open_to_namei_flags(op->open_flag);
2789 int create_error = 0;
2790 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2793 BUG_ON(dentry->d_inode);
2795 /* Don't create child dentry for a dead directory. */
2796 if (unlikely(IS_DEADDIR(dir))) {
2802 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2803 mode &= ~current_umask();
2805 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2807 open_flag &= ~O_TRUNC;
2810 * Checking write permission is tricky, bacuse we don't know if we are
2811 * going to actually need it: O_CREAT opens should work as long as the
2812 * file exists. But checking existence breaks atomicity. The trick is
2813 * to check access and if not granted clear O_CREAT from the flags.
2815 * Another problem is returing the "right" error value (e.g. for an
2816 * O_EXCL open we want to return EEXIST not EROFS).
2818 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2819 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2820 if (!(open_flag & O_CREAT)) {
2822 * No O_CREATE -> atomicity not a requirement -> fall
2823 * back to lookup + open
2826 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2827 /* Fall back and fail with the right error */
2828 create_error = -EROFS;
2831 /* No side effects, safe to clear O_CREAT */
2832 create_error = -EROFS;
2833 open_flag &= ~O_CREAT;
2837 if (open_flag & O_CREAT) {
2838 error = may_o_create(&nd->path, dentry, mode);
2840 create_error = error;
2841 if (open_flag & O_EXCL)
2843 open_flag &= ~O_CREAT;
2847 if (nd->flags & LOOKUP_DIRECTORY)
2848 open_flag |= O_DIRECTORY;
2850 file->f_path.dentry = DENTRY_NOT_SET;
2851 file->f_path.mnt = nd->path.mnt;
2852 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2855 if (create_error && error == -ENOENT)
2856 error = create_error;
2860 if (error) { /* returned 1, that is */
2861 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2865 if (file->f_path.dentry) {
2867 dentry = file->f_path.dentry;
2869 if (*opened & FILE_CREATED)
2870 fsnotify_create(dir, dentry);
2871 if (!dentry->d_inode) {
2872 WARN_ON(*opened & FILE_CREATED);
2874 error = create_error;
2878 if (excl && !(*opened & FILE_CREATED)) {
2887 * We didn't have the inode before the open, so check open permission
2890 acc_mode = op->acc_mode;
2891 if (*opened & FILE_CREATED) {
2892 WARN_ON(!(open_flag & O_CREAT));
2893 fsnotify_create(dir, dentry);
2894 acc_mode = MAY_OPEN;
2896 error = may_open(&file->f_path, acc_mode, open_flag);
2906 dentry = lookup_real(dir, dentry, nd->flags);
2908 return PTR_ERR(dentry);
2910 if (create_error && !dentry->d_inode) {
2911 error = create_error;
2915 path->dentry = dentry;
2916 path->mnt = nd->path.mnt;
2921 * Look up and maybe create and open the last component.
2923 * Must be called with i_mutex held on parent.
2925 * Returns 0 if the file was successfully atomically created (if necessary) and
2926 * opened. In this case the file will be returned attached to @file.
2928 * Returns 1 if the file was not completely opened at this time, though lookups
2929 * and creations will have been performed and the dentry returned in @path will
2930 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2931 * specified then a negative dentry may be returned.
2933 * An error code is returned otherwise.
2935 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2936 * cleared otherwise prior to returning.
2938 static int lookup_open(struct nameidata *nd, struct path *path,
2940 const struct open_flags *op,
2941 bool got_write, int *opened)
2943 struct dentry *dir = nd->path.dentry;
2944 struct inode *dir_inode = dir->d_inode;
2945 struct dentry *dentry;
2949 *opened &= ~FILE_CREATED;
2950 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2952 return PTR_ERR(dentry);
2954 /* Cached positive dentry: will open in f_op->open */
2955 if (!need_lookup && dentry->d_inode)
2958 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2959 return atomic_open(nd, dentry, path, file, op, got_write,
2960 need_lookup, opened);
2964 BUG_ON(dentry->d_inode);
2966 dentry = lookup_real(dir_inode, dentry, nd->flags);
2968 return PTR_ERR(dentry);
2971 /* Negative dentry, just create the file */
2972 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2973 umode_t mode = op->mode;
2974 if (!IS_POSIXACL(dir->d_inode))
2975 mode &= ~current_umask();
2977 * This write is needed to ensure that a
2978 * rw->ro transition does not occur between
2979 * the time when the file is created and when
2980 * a permanent write count is taken through
2981 * the 'struct file' in finish_open().
2987 *opened |= FILE_CREATED;
2988 error = security_path_mknod(&nd->path, dentry, mode, 0);
2991 error = vfs_create(dir->d_inode, dentry, mode,
2992 nd->flags & LOOKUP_EXCL);
2997 path->dentry = dentry;
2998 path->mnt = nd->path.mnt;
3007 * Handle the last step of open()
3009 static int do_last(struct nameidata *nd,
3010 struct file *file, const struct open_flags *op,
3013 struct dentry *dir = nd->path.dentry;
3014 int open_flag = op->open_flag;
3015 bool will_truncate = (open_flag & O_TRUNC) != 0;
3016 bool got_write = false;
3017 int acc_mode = op->acc_mode;
3019 struct inode *inode;
3020 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3022 bool retried = false;
3025 nd->flags &= ~LOOKUP_PARENT;
3026 nd->flags |= op->intent;
3028 if (nd->last_type != LAST_NORM) {
3029 error = handle_dots(nd, nd->last_type);
3030 if (unlikely(error))
3035 if (!(open_flag & O_CREAT)) {
3036 if (nd->last.name[nd->last.len])
3037 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3038 /* we _can_ be in RCU mode here */
3039 error = lookup_fast(nd, &path, &inode, &seq);
3046 BUG_ON(nd->inode != dir->d_inode);
3048 /* create side of things */
3050 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3051 * has been cleared when we got to the last component we are
3054 error = complete_walk(nd);
3058 audit_inode(nd->name, dir, LOOKUP_PARENT);
3059 /* trailing slashes? */
3060 if (unlikely(nd->last.name[nd->last.len]))
3065 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3066 error = mnt_want_write(nd->path.mnt);
3070 * do _not_ fail yet - we might not need that or fail with
3071 * a different error; let lookup_open() decide; we'll be
3072 * dropping this one anyway.
3075 mutex_lock(&dir->d_inode->i_mutex);
3076 error = lookup_open(nd, &path, file, op, got_write, opened);
3077 mutex_unlock(&dir->d_inode->i_mutex);
3083 if ((*opened & FILE_CREATED) ||
3084 !S_ISREG(file_inode(file)->i_mode))
3085 will_truncate = false;
3087 audit_inode(nd->name, file->f_path.dentry, 0);
3091 if (*opened & FILE_CREATED) {
3092 /* Don't check for write permission, don't truncate */
3093 open_flag &= ~O_TRUNC;
3094 will_truncate = false;
3095 acc_mode = MAY_OPEN;
3096 path_to_nameidata(&path, nd);
3097 goto finish_open_created;
3101 * create/update audit record if it already exists.
3103 if (d_is_positive(path.dentry))
3104 audit_inode(nd->name, path.dentry, 0);
3107 * If atomic_open() acquired write access it is dropped now due to
3108 * possible mount and symlink following (this might be optimized away if
3112 mnt_drop_write(nd->path.mnt);
3116 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3117 path_to_nameidata(&path, nd);
3121 error = follow_managed(&path, nd);
3122 if (unlikely(error < 0))
3125 BUG_ON(nd->flags & LOOKUP_RCU);
3126 seq = 0; /* out of RCU mode, so the value doesn't matter */
3127 if (unlikely(d_is_negative(path.dentry))) {
3128 path_to_nameidata(&path, nd);
3131 inode = d_backing_inode(path.dentry);
3135 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3137 if (unlikely(error))
3140 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3141 path_to_nameidata(&path, nd);
3143 save_parent.dentry = nd->path.dentry;
3144 save_parent.mnt = mntget(path.mnt);
3145 nd->path.dentry = path.dentry;
3150 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3152 error = complete_walk(nd);
3154 path_put(&save_parent);
3157 audit_inode(nd->name, nd->path.dentry, 0);
3158 if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3163 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3166 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3168 if (!d_is_reg(nd->path.dentry))
3169 will_truncate = false;
3171 if (will_truncate) {
3172 error = mnt_want_write(nd->path.mnt);
3177 finish_open_created:
3178 error = may_open(&nd->path, acc_mode, open_flag);
3182 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3183 error = vfs_open(&nd->path, file, current_cred());
3185 *opened |= FILE_OPENED;
3187 if (error == -EOPENSTALE)
3192 error = open_check_o_direct(file);
3195 error = ima_file_check(file, op->acc_mode, *opened);
3199 if (will_truncate) {
3200 error = handle_truncate(file);
3205 if (unlikely(error > 0)) {
3210 mnt_drop_write(nd->path.mnt);
3211 path_put(&save_parent);
3219 /* If no saved parent or already retried then can't retry */
3220 if (!save_parent.dentry || retried)
3223 BUG_ON(save_parent.dentry != dir);
3224 path_put(&nd->path);
3225 nd->path = save_parent;
3226 nd->inode = dir->d_inode;
3227 save_parent.mnt = NULL;
3228 save_parent.dentry = NULL;
3230 mnt_drop_write(nd->path.mnt);
3237 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3238 const struct open_flags *op,
3239 struct file *file, int *opened)
3241 static const struct qstr name = QSTR_INIT("/", 1);
3242 struct dentry *child;
3245 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3246 if (unlikely(error))
3248 error = mnt_want_write(path.mnt);
3249 if (unlikely(error))
3251 dir = path.dentry->d_inode;
3252 /* we want directory to be writable */
3253 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3256 if (!dir->i_op->tmpfile) {
3257 error = -EOPNOTSUPP;
3260 child = d_alloc(path.dentry, &name);
3261 if (unlikely(!child)) {
3266 path.dentry = child;
3267 error = dir->i_op->tmpfile(dir, child, op->mode);
3270 audit_inode(nd->name, child, 0);
3271 /* Don't check for other permissions, the inode was just created */
3272 error = may_open(&path, MAY_OPEN, op->open_flag);
3275 file->f_path.mnt = path.mnt;
3276 error = finish_open(file, child, NULL, opened);
3279 error = open_check_o_direct(file);
3282 } else if (!(op->open_flag & O_EXCL)) {
3283 struct inode *inode = file_inode(file);
3284 spin_lock(&inode->i_lock);
3285 inode->i_state |= I_LINKABLE;
3286 spin_unlock(&inode->i_lock);
3289 mnt_drop_write(path.mnt);
3295 static struct file *path_openat(struct nameidata *nd,
3296 const struct open_flags *op, unsigned flags)
3303 file = get_empty_filp();
3307 file->f_flags = op->open_flag;
3309 if (unlikely(file->f_flags & __O_TMPFILE)) {
3310 error = do_tmpfile(nd, flags, op, file, &opened);
3314 s = path_init(nd, flags);
3319 while (!(error = link_path_walk(s, nd)) &&
3320 (error = do_last(nd, file, op, &opened)) > 0) {
3321 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3322 s = trailing_symlink(nd);
3330 if (!(opened & FILE_OPENED)) {
3334 if (unlikely(error)) {
3335 if (error == -EOPENSTALE) {
3336 if (flags & LOOKUP_RCU)
3341 file = ERR_PTR(error);
3346 struct file *do_filp_open(int dfd, struct filename *pathname,
3347 const struct open_flags *op)
3349 struct nameidata nd;
3350 int flags = op->lookup_flags;
3353 set_nameidata(&nd, dfd, pathname);
3354 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3355 if (unlikely(filp == ERR_PTR(-ECHILD)))
3356 filp = path_openat(&nd, op, flags);
3357 if (unlikely(filp == ERR_PTR(-ESTALE)))
3358 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3359 restore_nameidata();
3363 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3364 const char *name, const struct open_flags *op)
3366 struct nameidata nd;
3368 struct filename *filename;
3369 int flags = op->lookup_flags | LOOKUP_ROOT;
3372 nd.root.dentry = dentry;
3374 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3375 return ERR_PTR(-ELOOP);
3377 filename = getname_kernel(name);
3378 if (IS_ERR(filename))
3379 return ERR_CAST(filename);
3381 set_nameidata(&nd, -1, filename);
3382 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3383 if (unlikely(file == ERR_PTR(-ECHILD)))
3384 file = path_openat(&nd, op, flags);
3385 if (unlikely(file == ERR_PTR(-ESTALE)))
3386 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3387 restore_nameidata();
3392 static struct dentry *filename_create(int dfd, struct filename *name,
3393 struct path *path, unsigned int lookup_flags)
3395 struct dentry *dentry = ERR_PTR(-EEXIST);
3400 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3403 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3404 * other flags passed in are ignored!
3406 lookup_flags &= LOOKUP_REVAL;
3408 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3410 return ERR_CAST(name);
3413 * Yucky last component or no last component at all?
3414 * (foo/., foo/.., /////)
3416 if (unlikely(type != LAST_NORM))
3419 /* don't fail immediately if it's r/o, at least try to report other errors */
3420 err2 = mnt_want_write(path->mnt);
3422 * Do the final lookup.
3424 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3425 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3426 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3431 if (d_is_positive(dentry))
3435 * Special case - lookup gave negative, but... we had foo/bar/
3436 * From the vfs_mknod() POV we just have a negative dentry -
3437 * all is fine. Let's be bastards - you had / on the end, you've
3438 * been asking for (non-existent) directory. -ENOENT for you.
3440 if (unlikely(!is_dir && last.name[last.len])) {
3444 if (unlikely(err2)) {
3452 dentry = ERR_PTR(error);
3454 mutex_unlock(&path->dentry->d_inode->i_mutex);
3456 mnt_drop_write(path->mnt);
3463 struct dentry *kern_path_create(int dfd, const char *pathname,
3464 struct path *path, unsigned int lookup_flags)
3466 return filename_create(dfd, getname_kernel(pathname),
3467 path, lookup_flags);
3469 EXPORT_SYMBOL(kern_path_create);
3471 void done_path_create(struct path *path, struct dentry *dentry)
3474 mutex_unlock(&path->dentry->d_inode->i_mutex);
3475 mnt_drop_write(path->mnt);
3478 EXPORT_SYMBOL(done_path_create);
3480 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3481 struct path *path, unsigned int lookup_flags)
3483 return filename_create(dfd, getname(pathname), path, lookup_flags);
3485 EXPORT_SYMBOL(user_path_create);
3487 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3489 int error = may_create(dir, dentry);
3494 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3497 if (!dir->i_op->mknod)
3500 error = devcgroup_inode_mknod(mode, dev);
3504 error = security_inode_mknod(dir, dentry, mode, dev);
3508 error = dir->i_op->mknod(dir, dentry, mode, dev);
3510 fsnotify_create(dir, dentry);
3513 EXPORT_SYMBOL(vfs_mknod);
3515 static int may_mknod(umode_t mode)
3517 switch (mode & S_IFMT) {
3523 case 0: /* zero mode translates to S_IFREG */
3532 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3535 struct dentry *dentry;
3538 unsigned int lookup_flags = 0;
3540 error = may_mknod(mode);
3544 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3546 return PTR_ERR(dentry);
3548 if (!IS_POSIXACL(path.dentry->d_inode))
3549 mode &= ~current_umask();
3550 error = security_path_mknod(&path, dentry, mode, dev);
3553 switch (mode & S_IFMT) {
3554 case 0: case S_IFREG:
3555 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3557 case S_IFCHR: case S_IFBLK:
3558 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3559 new_decode_dev(dev));
3561 case S_IFIFO: case S_IFSOCK:
3562 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3566 done_path_create(&path, dentry);
3567 if (retry_estale(error, lookup_flags)) {
3568 lookup_flags |= LOOKUP_REVAL;
3574 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3576 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3579 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3581 int error = may_create(dir, dentry);
3582 unsigned max_links = dir->i_sb->s_max_links;
3587 if (!dir->i_op->mkdir)
3590 mode &= (S_IRWXUGO|S_ISVTX);
3591 error = security_inode_mkdir(dir, dentry, mode);
3595 if (max_links && dir->i_nlink >= max_links)
3598 error = dir->i_op->mkdir(dir, dentry, mode);
3600 fsnotify_mkdir(dir, dentry);
3603 EXPORT_SYMBOL(vfs_mkdir);
3605 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3607 struct dentry *dentry;
3610 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3613 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3615 return PTR_ERR(dentry);
3617 if (!IS_POSIXACL(path.dentry->d_inode))
3618 mode &= ~current_umask();
3619 error = security_path_mkdir(&path, dentry, mode);
3621 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3622 done_path_create(&path, dentry);
3623 if (retry_estale(error, lookup_flags)) {
3624 lookup_flags |= LOOKUP_REVAL;
3630 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3632 return sys_mkdirat(AT_FDCWD, pathname, mode);
3636 * The dentry_unhash() helper will try to drop the dentry early: we
3637 * should have a usage count of 1 if we're the only user of this
3638 * dentry, and if that is true (possibly after pruning the dcache),
3639 * then we drop the dentry now.
3641 * A low-level filesystem can, if it choses, legally
3644 * if (!d_unhashed(dentry))
3647 * if it cannot handle the case of removing a directory
3648 * that is still in use by something else..
3650 void dentry_unhash(struct dentry *dentry)
3652 shrink_dcache_parent(dentry);
3653 spin_lock(&dentry->d_lock);
3654 if (dentry->d_lockref.count == 1)
3656 spin_unlock(&dentry->d_lock);
3658 EXPORT_SYMBOL(dentry_unhash);
3660 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3662 int error = may_delete(dir, dentry, 1);
3667 if (!dir->i_op->rmdir)
3671 mutex_lock(&dentry->d_inode->i_mutex);
3674 if (is_local_mountpoint(dentry))
3677 error = security_inode_rmdir(dir, dentry);
3681 shrink_dcache_parent(dentry);
3682 error = dir->i_op->rmdir(dir, dentry);
3686 dentry->d_inode->i_flags |= S_DEAD;
3688 detach_mounts(dentry);
3691 mutex_unlock(&dentry->d_inode->i_mutex);
3697 EXPORT_SYMBOL(vfs_rmdir);
3699 static long do_rmdir(int dfd, const char __user *pathname)
3702 struct filename *name;
3703 struct dentry *dentry;
3707 unsigned int lookup_flags = 0;
3709 name = user_path_parent(dfd, pathname,
3710 &path, &last, &type, lookup_flags);
3712 return PTR_ERR(name);
3726 error = mnt_want_write(path.mnt);
3730 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3731 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3732 error = PTR_ERR(dentry);
3735 if (!dentry->d_inode) {
3739 error = security_path_rmdir(&path, dentry);
3742 error = vfs_rmdir(path.dentry->d_inode, dentry);
3746 mutex_unlock(&path.dentry->d_inode->i_mutex);
3747 mnt_drop_write(path.mnt);
3751 if (retry_estale(error, lookup_flags)) {
3752 lookup_flags |= LOOKUP_REVAL;
3758 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3760 return do_rmdir(AT_FDCWD, pathname);
3764 * vfs_unlink - unlink a filesystem object
3765 * @dir: parent directory
3767 * @delegated_inode: returns victim inode, if the inode is delegated.
3769 * The caller must hold dir->i_mutex.
3771 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3772 * return a reference to the inode in delegated_inode. The caller
3773 * should then break the delegation on that inode and retry. Because
3774 * breaking a delegation may take a long time, the caller should drop
3775 * dir->i_mutex before doing so.
3777 * Alternatively, a caller may pass NULL for delegated_inode. This may
3778 * be appropriate for callers that expect the underlying filesystem not
3779 * to be NFS exported.
3781 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3783 struct inode *target = dentry->d_inode;
3784 int error = may_delete(dir, dentry, 0);
3789 if (!dir->i_op->unlink)
3792 mutex_lock(&target->i_mutex);
3793 if (is_local_mountpoint(dentry))
3796 error = security_inode_unlink(dir, dentry);
3798 error = try_break_deleg(target, delegated_inode);
3801 error = dir->i_op->unlink(dir, dentry);
3804 detach_mounts(dentry);
3809 mutex_unlock(&target->i_mutex);
3811 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3812 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3813 fsnotify_link_count(target);
3819 EXPORT_SYMBOL(vfs_unlink);
3822 * Make sure that the actual truncation of the file will occur outside its
3823 * directory's i_mutex. Truncate can take a long time if there is a lot of
3824 * writeout happening, and we don't want to prevent access to the directory
3825 * while waiting on the I/O.
3827 static long do_unlinkat(int dfd, const char __user *pathname)
3830 struct filename *name;
3831 struct dentry *dentry;
3835 struct inode *inode = NULL;
3836 struct inode *delegated_inode = NULL;
3837 unsigned int lookup_flags = 0;
3839 name = user_path_parent(dfd, pathname,
3840 &path, &last, &type, lookup_flags);
3842 return PTR_ERR(name);
3845 if (type != LAST_NORM)
3848 error = mnt_want_write(path.mnt);
3852 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3853 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3854 error = PTR_ERR(dentry);
3855 if (!IS_ERR(dentry)) {
3856 /* Why not before? Because we want correct error value */
3857 if (last.name[last.len])
3859 inode = dentry->d_inode;
3860 if (d_is_negative(dentry))
3863 error = security_path_unlink(&path, dentry);
3866 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3870 mutex_unlock(&path.dentry->d_inode->i_mutex);
3872 iput(inode); /* truncate the inode here */
3874 if (delegated_inode) {
3875 error = break_deleg_wait(&delegated_inode);
3879 mnt_drop_write(path.mnt);
3883 if (retry_estale(error, lookup_flags)) {
3884 lookup_flags |= LOOKUP_REVAL;
3891 if (d_is_negative(dentry))
3893 else if (d_is_dir(dentry))
3900 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3902 if ((flag & ~AT_REMOVEDIR) != 0)
3905 if (flag & AT_REMOVEDIR)
3906 return do_rmdir(dfd, pathname);
3908 return do_unlinkat(dfd, pathname);
3911 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3913 return do_unlinkat(AT_FDCWD, pathname);
3916 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3918 int error = may_create(dir, dentry);
3923 if (!dir->i_op->symlink)
3926 error = security_inode_symlink(dir, dentry, oldname);
3930 error = dir->i_op->symlink(dir, dentry, oldname);
3932 fsnotify_create(dir, dentry);
3935 EXPORT_SYMBOL(vfs_symlink);
3937 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3938 int, newdfd, const char __user *, newname)
3941 struct filename *from;
3942 struct dentry *dentry;
3944 unsigned int lookup_flags = 0;
3946 from = getname(oldname);
3948 return PTR_ERR(from);
3950 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3951 error = PTR_ERR(dentry);
3955 error = security_path_symlink(&path, dentry, from->name);
3957 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3958 done_path_create(&path, dentry);
3959 if (retry_estale(error, lookup_flags)) {
3960 lookup_flags |= LOOKUP_REVAL;
3968 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3970 return sys_symlinkat(oldname, AT_FDCWD, newname);
3974 * vfs_link - create a new link
3975 * @old_dentry: object to be linked
3977 * @new_dentry: where to create the new link
3978 * @delegated_inode: returns inode needing a delegation break
3980 * The caller must hold dir->i_mutex
3982 * If vfs_link discovers a delegation on the to-be-linked file in need
3983 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3984 * inode in delegated_inode. The caller should then break the delegation
3985 * and retry. Because breaking a delegation may take a long time, the
3986 * caller should drop the i_mutex before doing so.
3988 * Alternatively, a caller may pass NULL for delegated_inode. This may
3989 * be appropriate for callers that expect the underlying filesystem not
3990 * to be NFS exported.
3992 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3994 struct inode *inode = old_dentry->d_inode;
3995 unsigned max_links = dir->i_sb->s_max_links;
4001 error = may_create(dir, new_dentry);
4005 if (dir->i_sb != inode->i_sb)
4009 * A link to an append-only or immutable file cannot be created.
4011 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4013 if (!dir->i_op->link)
4015 if (S_ISDIR(inode->i_mode))
4018 error = security_inode_link(old_dentry, dir, new_dentry);
4022 mutex_lock(&inode->i_mutex);
4023 /* Make sure we don't allow creating hardlink to an unlinked file */
4024 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4026 else if (max_links && inode->i_nlink >= max_links)
4029 error = try_break_deleg(inode, delegated_inode);
4031 error = dir->i_op->link(old_dentry, dir, new_dentry);
4034 if (!error && (inode->i_state & I_LINKABLE)) {
4035 spin_lock(&inode->i_lock);
4036 inode->i_state &= ~I_LINKABLE;
4037 spin_unlock(&inode->i_lock);
4039 mutex_unlock(&inode->i_mutex);
4041 fsnotify_link(dir, inode, new_dentry);
4044 EXPORT_SYMBOL(vfs_link);
4047 * Hardlinks are often used in delicate situations. We avoid
4048 * security-related surprises by not following symlinks on the
4051 * We don't follow them on the oldname either to be compatible
4052 * with linux 2.0, and to avoid hard-linking to directories
4053 * and other special files. --ADM
4055 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4056 int, newdfd, const char __user *, newname, int, flags)
4058 struct dentry *new_dentry;
4059 struct path old_path, new_path;
4060 struct inode *delegated_inode = NULL;
4064 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4067 * To use null names we require CAP_DAC_READ_SEARCH
4068 * This ensures that not everyone will be able to create
4069 * handlink using the passed filedescriptor.
4071 if (flags & AT_EMPTY_PATH) {
4072 if (!capable(CAP_DAC_READ_SEARCH))
4077 if (flags & AT_SYMLINK_FOLLOW)
4078 how |= LOOKUP_FOLLOW;
4080 error = user_path_at(olddfd, oldname, how, &old_path);
4084 new_dentry = user_path_create(newdfd, newname, &new_path,
4085 (how & LOOKUP_REVAL));
4086 error = PTR_ERR(new_dentry);
4087 if (IS_ERR(new_dentry))
4091 if (old_path.mnt != new_path.mnt)
4093 error = may_linkat(&old_path);
4094 if (unlikely(error))
4096 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4099 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4101 done_path_create(&new_path, new_dentry);
4102 if (delegated_inode) {
4103 error = break_deleg_wait(&delegated_inode);
4105 path_put(&old_path);
4109 if (retry_estale(error, how)) {
4110 path_put(&old_path);
4111 how |= LOOKUP_REVAL;
4115 path_put(&old_path);
4120 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4122 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4126 * vfs_rename - rename a filesystem object
4127 * @old_dir: parent of source
4128 * @old_dentry: source
4129 * @new_dir: parent of destination
4130 * @new_dentry: destination
4131 * @delegated_inode: returns an inode needing a delegation break
4132 * @flags: rename flags
4134 * The caller must hold multiple mutexes--see lock_rename()).
4136 * If vfs_rename discovers a delegation in need of breaking at either
4137 * the source or destination, it will return -EWOULDBLOCK and return a
4138 * reference to the inode in delegated_inode. The caller should then
4139 * break the delegation and retry. Because breaking a delegation may
4140 * take a long time, the caller should drop all locks before doing
4143 * Alternatively, a caller may pass NULL for delegated_inode. This may
4144 * be appropriate for callers that expect the underlying filesystem not
4145 * to be NFS exported.
4147 * The worst of all namespace operations - renaming directory. "Perverted"
4148 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4150 * a) we can get into loop creation.
4151 * b) race potential - two innocent renames can create a loop together.
4152 * That's where 4.4 screws up. Current fix: serialization on
4153 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4155 * c) we have to lock _four_ objects - parents and victim (if it exists),
4156 * and source (if it is not a directory).
4157 * And that - after we got ->i_mutex on parents (until then we don't know
4158 * whether the target exists). Solution: try to be smart with locking
4159 * order for inodes. We rely on the fact that tree topology may change
4160 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4161 * move will be locked. Thus we can rank directories by the tree
4162 * (ancestors first) and rank all non-directories after them.
4163 * That works since everybody except rename does "lock parent, lookup,
4164 * lock child" and rename is under ->s_vfs_rename_mutex.
4165 * HOWEVER, it relies on the assumption that any object with ->lookup()
4166 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4167 * we'd better make sure that there's no link(2) for them.
4168 * d) conversion from fhandle to dentry may come in the wrong moment - when
4169 * we are removing the target. Solution: we will have to grab ->i_mutex
4170 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4171 * ->i_mutex on parents, which works but leads to some truly excessive
4174 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4175 struct inode *new_dir, struct dentry *new_dentry,
4176 struct inode **delegated_inode, unsigned int flags)
4179 bool is_dir = d_is_dir(old_dentry);
4180 const unsigned char *old_name;
4181 struct inode *source = old_dentry->d_inode;
4182 struct inode *target = new_dentry->d_inode;
4183 bool new_is_dir = false;
4184 unsigned max_links = new_dir->i_sb->s_max_links;
4187 * Check source == target.
4188 * On overlayfs need to look at underlying inodes.
4190 if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4193 error = may_delete(old_dir, old_dentry, is_dir);
4198 error = may_create(new_dir, new_dentry);
4200 new_is_dir = d_is_dir(new_dentry);
4202 if (!(flags & RENAME_EXCHANGE))
4203 error = may_delete(new_dir, new_dentry, is_dir);
4205 error = may_delete(new_dir, new_dentry, new_is_dir);
4210 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4213 if (flags && !old_dir->i_op->rename2)
4217 * If we are going to change the parent - check write permissions,
4218 * we'll need to flip '..'.
4220 if (new_dir != old_dir) {
4222 error = inode_permission(source, MAY_WRITE);
4226 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4227 error = inode_permission(target, MAY_WRITE);
4233 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4238 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4240 if (!is_dir || (flags & RENAME_EXCHANGE))
4241 lock_two_nondirectories(source, target);
4243 mutex_lock(&target->i_mutex);
4246 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4249 if (max_links && new_dir != old_dir) {
4251 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4253 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4254 old_dir->i_nlink >= max_links)
4257 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4258 shrink_dcache_parent(new_dentry);
4260 error = try_break_deleg(source, delegated_inode);
4264 if (target && !new_is_dir) {
4265 error = try_break_deleg(target, delegated_inode);
4269 if (!old_dir->i_op->rename2) {
4270 error = old_dir->i_op->rename(old_dir, old_dentry,
4271 new_dir, new_dentry);
4273 WARN_ON(old_dir->i_op->rename != NULL);
4274 error = old_dir->i_op->rename2(old_dir, old_dentry,
4275 new_dir, new_dentry, flags);
4280 if (!(flags & RENAME_EXCHANGE) && target) {
4282 target->i_flags |= S_DEAD;
4283 dont_mount(new_dentry);
4284 detach_mounts(new_dentry);
4286 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4287 if (!(flags & RENAME_EXCHANGE))
4288 d_move(old_dentry, new_dentry);
4290 d_exchange(old_dentry, new_dentry);
4293 if (!is_dir || (flags & RENAME_EXCHANGE))
4294 unlock_two_nondirectories(source, target);
4296 mutex_unlock(&target->i_mutex);
4299 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4300 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4301 if (flags & RENAME_EXCHANGE) {
4302 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4303 new_is_dir, NULL, new_dentry);
4306 fsnotify_oldname_free(old_name);
4310 EXPORT_SYMBOL(vfs_rename);
4312 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4313 int, newdfd, const char __user *, newname, unsigned int, flags)
4315 struct dentry *old_dentry, *new_dentry;
4316 struct dentry *trap;
4317 struct path old_path, new_path;
4318 struct qstr old_last, new_last;
4319 int old_type, new_type;
4320 struct inode *delegated_inode = NULL;
4321 struct filename *from;
4322 struct filename *to;
4323 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4324 bool should_retry = false;
4327 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4330 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4331 (flags & RENAME_EXCHANGE))
4334 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4337 if (flags & RENAME_EXCHANGE)
4341 from = user_path_parent(olddfd, oldname,
4342 &old_path, &old_last, &old_type, lookup_flags);
4344 error = PTR_ERR(from);
4348 to = user_path_parent(newdfd, newname,
4349 &new_path, &new_last, &new_type, lookup_flags);
4351 error = PTR_ERR(to);
4356 if (old_path.mnt != new_path.mnt)
4360 if (old_type != LAST_NORM)
4363 if (flags & RENAME_NOREPLACE)
4365 if (new_type != LAST_NORM)
4368 error = mnt_want_write(old_path.mnt);
4373 trap = lock_rename(new_path.dentry, old_path.dentry);
4375 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4376 error = PTR_ERR(old_dentry);
4377 if (IS_ERR(old_dentry))
4379 /* source must exist */
4381 if (d_is_negative(old_dentry))
4383 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4384 error = PTR_ERR(new_dentry);
4385 if (IS_ERR(new_dentry))
4388 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4390 if (flags & RENAME_EXCHANGE) {
4392 if (d_is_negative(new_dentry))
4395 if (!d_is_dir(new_dentry)) {
4397 if (new_last.name[new_last.len])
4401 /* unless the source is a directory trailing slashes give -ENOTDIR */
4402 if (!d_is_dir(old_dentry)) {
4404 if (old_last.name[old_last.len])
4406 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4409 /* source should not be ancestor of target */
4411 if (old_dentry == trap)
4413 /* target should not be an ancestor of source */
4414 if (!(flags & RENAME_EXCHANGE))
4416 if (new_dentry == trap)
4419 error = security_path_rename(&old_path, old_dentry,
4420 &new_path, new_dentry, flags);
4423 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4424 new_path.dentry->d_inode, new_dentry,
4425 &delegated_inode, flags);
4431 unlock_rename(new_path.dentry, old_path.dentry);
4432 if (delegated_inode) {
4433 error = break_deleg_wait(&delegated_inode);
4437 mnt_drop_write(old_path.mnt);
4439 if (retry_estale(error, lookup_flags))
4440 should_retry = true;
4441 path_put(&new_path);
4444 path_put(&old_path);
4447 should_retry = false;
4448 lookup_flags |= LOOKUP_REVAL;
4455 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4456 int, newdfd, const char __user *, newname)
4458 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4461 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4463 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4466 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4468 int error = may_create(dir, dentry);
4472 if (!dir->i_op->mknod)
4475 return dir->i_op->mknod(dir, dentry,
4476 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4478 EXPORT_SYMBOL(vfs_whiteout);
4480 int readlink_copy(char __user *buffer, int buflen, const char *link)
4482 int len = PTR_ERR(link);
4487 if (len > (unsigned) buflen)
4489 if (copy_to_user(buffer, link, len))
4494 EXPORT_SYMBOL(readlink_copy);
4497 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4498 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4499 * using) it for any given inode is up to filesystem.
4501 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4504 struct inode *inode = d_inode(dentry);
4505 const char *link = inode->i_link;
4509 link = inode->i_op->follow_link(dentry, &cookie);
4511 return PTR_ERR(link);
4513 res = readlink_copy(buffer, buflen, link);
4514 if (inode->i_op->put_link)
4515 inode->i_op->put_link(inode, cookie);
4518 EXPORT_SYMBOL(generic_readlink);
4520 /* get the link contents into pagecache */
4521 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4525 struct address_space *mapping = dentry->d_inode->i_mapping;
4526 page = read_mapping_page(mapping, 0, NULL);
4531 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4535 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4537 struct page *page = NULL;
4538 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4541 page_cache_release(page);
4545 EXPORT_SYMBOL(page_readlink);
4547 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4549 struct page *page = NULL;
4550 char *res = page_getlink(dentry, &page);
4555 EXPORT_SYMBOL(page_follow_link_light);
4557 void page_put_link(struct inode *unused, void *cookie)
4559 struct page *page = cookie;
4561 page_cache_release(page);
4563 EXPORT_SYMBOL(page_put_link);
4566 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4568 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4570 struct address_space *mapping = inode->i_mapping;
4575 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4577 flags |= AOP_FLAG_NOFS;
4580 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4581 flags, &page, &fsdata);
4585 kaddr = kmap_atomic(page);
4586 memcpy(kaddr, symname, len-1);
4587 kunmap_atomic(kaddr);
4589 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4596 mark_inode_dirty(inode);
4601 EXPORT_SYMBOL(__page_symlink);
4603 int page_symlink(struct inode *inode, const char *symname, int len)
4605 return __page_symlink(inode, symname, len,
4606 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4608 EXPORT_SYMBOL(page_symlink);
4610 const struct inode_operations page_symlink_inode_operations = {
4611 .readlink = generic_readlink,
4612 .follow_link = page_follow_link_light,
4613 .put_link = page_put_link,
4615 EXPORT_SYMBOL(page_symlink_inode_operations);