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))
1562 * This sequence count validates that the parent had no
1563 * changes while we did the lookup of the dentry above.
1565 * The memory barrier in read_seqcount_begin of child is
1566 * enough, we can use __read_seqcount_retry here.
1568 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1572 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1573 status = d_revalidate(dentry, nd->flags);
1574 if (unlikely(status <= 0)) {
1575 if (status != -ECHILD)
1581 path->dentry = dentry;
1582 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1585 if (unlazy_walk(nd, dentry, seq))
1588 dentry = __d_lookup(parent, &nd->last);
1591 if (unlikely(!dentry))
1594 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1595 status = d_revalidate(dentry, nd->flags);
1596 if (unlikely(status <= 0)) {
1601 d_invalidate(dentry);
1606 if (unlikely(d_is_negative(dentry))) {
1611 path->dentry = dentry;
1612 err = follow_managed(path, nd);
1614 *inode = d_backing_inode(path->dentry);
1621 /* Fast lookup failed, do it the slow way */
1622 static int lookup_slow(struct nameidata *nd, struct path *path)
1624 struct dentry *dentry, *parent;
1626 parent = nd->path.dentry;
1627 BUG_ON(nd->inode != parent->d_inode);
1629 mutex_lock(&parent->d_inode->i_mutex);
1630 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1631 mutex_unlock(&parent->d_inode->i_mutex);
1633 return PTR_ERR(dentry);
1634 path->mnt = nd->path.mnt;
1635 path->dentry = dentry;
1636 return follow_managed(path, nd);
1639 static inline int may_lookup(struct nameidata *nd)
1641 if (nd->flags & LOOKUP_RCU) {
1642 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1645 if (unlazy_walk(nd, NULL, 0))
1648 return inode_permission(nd->inode, MAY_EXEC);
1651 static inline int handle_dots(struct nameidata *nd, int type)
1653 if (type == LAST_DOTDOT) {
1654 if (nd->flags & LOOKUP_RCU) {
1655 return follow_dotdot_rcu(nd);
1657 return follow_dotdot(nd);
1662 static int pick_link(struct nameidata *nd, struct path *link,
1663 struct inode *inode, unsigned seq)
1667 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1668 path_to_nameidata(link, nd);
1671 if (!(nd->flags & LOOKUP_RCU)) {
1672 if (link->mnt == nd->path.mnt)
1675 error = nd_alloc_stack(nd);
1676 if (unlikely(error)) {
1677 if (error == -ECHILD) {
1678 if (unlikely(unlazy_link(nd, link, seq)))
1680 error = nd_alloc_stack(nd);
1688 last = nd->stack + nd->depth++;
1690 last->cookie = NULL;
1691 last->inode = inode;
1697 * Do we need to follow links? We _really_ want to be able
1698 * to do this check without having to look at inode->i_op,
1699 * so we keep a cache of "no, this doesn't need follow_link"
1700 * for the common case.
1702 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1704 struct inode *inode, unsigned seq)
1706 if (likely(!d_is_symlink(link->dentry)))
1710 return pick_link(nd, link, inode, seq);
1713 enum {WALK_GET = 1, WALK_PUT = 2};
1715 static int walk_component(struct nameidata *nd, int flags)
1718 struct inode *inode;
1722 * "." and ".." are special - ".." especially so because it has
1723 * to be able to know about the current root directory and
1724 * parent relationships.
1726 if (unlikely(nd->last_type != LAST_NORM)) {
1727 err = handle_dots(nd, nd->last_type);
1728 if (flags & WALK_PUT)
1732 err = lookup_fast(nd, &path, &inode, &seq);
1733 if (unlikely(err)) {
1737 err = lookup_slow(nd, &path);
1741 inode = d_backing_inode(path.dentry);
1742 seq = 0; /* we are already out of RCU mode */
1744 if (d_is_negative(path.dentry))
1748 if (flags & WALK_PUT)
1750 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1753 path_to_nameidata(&path, nd);
1759 path_to_nameidata(&path, nd);
1764 * We can do the critical dentry name comparison and hashing
1765 * operations one word at a time, but we are limited to:
1767 * - Architectures with fast unaligned word accesses. We could
1768 * do a "get_unaligned()" if this helps and is sufficiently
1771 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1772 * do not trap on the (extremely unlikely) case of a page
1773 * crossing operation.
1775 * - Furthermore, we need an efficient 64-bit compile for the
1776 * 64-bit case in order to generate the "number of bytes in
1777 * the final mask". Again, that could be replaced with a
1778 * efficient population count instruction or similar.
1780 #ifdef CONFIG_DCACHE_WORD_ACCESS
1782 #include <asm/word-at-a-time.h>
1786 static inline unsigned int fold_hash(unsigned long hash)
1788 return hash_64(hash, 32);
1791 #else /* 32-bit case */
1793 #define fold_hash(x) (x)
1797 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1799 unsigned long a, mask;
1800 unsigned long hash = 0;
1803 a = load_unaligned_zeropad(name);
1804 if (len < sizeof(unsigned long))
1808 name += sizeof(unsigned long);
1809 len -= sizeof(unsigned long);
1813 mask = bytemask_from_count(len);
1816 return fold_hash(hash);
1818 EXPORT_SYMBOL(full_name_hash);
1821 * Calculate the length and hash of the path component, and
1822 * return the "hash_len" as the result.
1824 static inline u64 hash_name(const char *name)
1826 unsigned long a, b, adata, bdata, mask, hash, len;
1827 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1830 len = -sizeof(unsigned long);
1832 hash = (hash + a) * 9;
1833 len += sizeof(unsigned long);
1834 a = load_unaligned_zeropad(name+len);
1835 b = a ^ REPEAT_BYTE('/');
1836 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1838 adata = prep_zero_mask(a, adata, &constants);
1839 bdata = prep_zero_mask(b, bdata, &constants);
1841 mask = create_zero_mask(adata | bdata);
1843 hash += a & zero_bytemask(mask);
1844 len += find_zero(mask);
1845 return hashlen_create(fold_hash(hash), len);
1850 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1852 unsigned long hash = init_name_hash();
1854 hash = partial_name_hash(*name++, hash);
1855 return end_name_hash(hash);
1857 EXPORT_SYMBOL(full_name_hash);
1860 * We know there's a real path component here of at least
1863 static inline u64 hash_name(const char *name)
1865 unsigned long hash = init_name_hash();
1866 unsigned long len = 0, c;
1868 c = (unsigned char)*name;
1871 hash = partial_name_hash(c, hash);
1872 c = (unsigned char)name[len];
1873 } while (c && c != '/');
1874 return hashlen_create(end_name_hash(hash), len);
1881 * This is the basic name resolution function, turning a pathname into
1882 * the final dentry. We expect 'base' to be positive and a directory.
1884 * Returns 0 and nd will have valid dentry and mnt on success.
1885 * Returns error and drops reference to input namei data on failure.
1887 static int link_path_walk(const char *name, struct nameidata *nd)
1896 /* At this point we know we have a real path component. */
1901 err = may_lookup(nd);
1905 hash_len = hash_name(name);
1908 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1910 if (name[1] == '.') {
1912 nd->flags |= LOOKUP_JUMPED;
1918 if (likely(type == LAST_NORM)) {
1919 struct dentry *parent = nd->path.dentry;
1920 nd->flags &= ~LOOKUP_JUMPED;
1921 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1922 struct qstr this = { { .hash_len = hash_len }, .name = name };
1923 err = parent->d_op->d_hash(parent, &this);
1926 hash_len = this.hash_len;
1931 nd->last.hash_len = hash_len;
1932 nd->last.name = name;
1933 nd->last_type = type;
1935 name += hashlen_len(hash_len);
1939 * If it wasn't NUL, we know it was '/'. Skip that
1940 * slash, and continue until no more slashes.
1944 } while (unlikely(*name == '/'));
1945 if (unlikely(!*name)) {
1947 /* pathname body, done */
1950 name = nd->stack[nd->depth - 1].name;
1951 /* trailing symlink, done */
1954 /* last component of nested symlink */
1955 err = walk_component(nd, WALK_GET | WALK_PUT);
1957 err = walk_component(nd, WALK_GET);
1963 const char *s = get_link(nd);
1965 if (unlikely(IS_ERR(s)))
1972 nd->stack[nd->depth - 1].name = name;
1977 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1978 if (nd->flags & LOOKUP_RCU) {
1979 if (unlazy_walk(nd, NULL, 0))
1987 static const char *path_init(struct nameidata *nd, unsigned flags)
1990 const char *s = nd->name->name;
1992 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1993 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1995 nd->total_link_count = 0;
1996 if (flags & LOOKUP_ROOT) {
1997 struct dentry *root = nd->root.dentry;
1998 struct inode *inode = root->d_inode;
2000 if (!d_can_lookup(root))
2001 return ERR_PTR(-ENOTDIR);
2002 retval = inode_permission(inode, MAY_EXEC);
2004 return ERR_PTR(retval);
2006 nd->path = nd->root;
2008 if (flags & LOOKUP_RCU) {
2010 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2011 nd->root_seq = nd->seq;
2012 nd->m_seq = read_seqbegin(&mount_lock);
2014 path_get(&nd->path);
2019 nd->root.mnt = NULL;
2021 nd->m_seq = read_seqbegin(&mount_lock);
2023 if (flags & LOOKUP_RCU) {
2026 nd->seq = nd->root_seq;
2029 path_get(&nd->root);
2031 nd->path = nd->root;
2032 } else if (nd->dfd == AT_FDCWD) {
2033 if (flags & LOOKUP_RCU) {
2034 struct fs_struct *fs = current->fs;
2040 seq = read_seqcount_begin(&fs->seq);
2042 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2043 } while (read_seqcount_retry(&fs->seq, seq));
2045 get_fs_pwd(current->fs, &nd->path);
2048 /* Caller must check execute permissions on the starting path component */
2049 struct fd f = fdget_raw(nd->dfd);
2050 struct dentry *dentry;
2053 return ERR_PTR(-EBADF);
2055 dentry = f.file->f_path.dentry;
2058 if (!d_can_lookup(dentry)) {
2060 return ERR_PTR(-ENOTDIR);
2064 nd->path = f.file->f_path;
2065 if (flags & LOOKUP_RCU) {
2067 nd->inode = nd->path.dentry->d_inode;
2068 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2070 path_get(&nd->path);
2071 nd->inode = nd->path.dentry->d_inode;
2077 nd->inode = nd->path.dentry->d_inode;
2078 if (!(flags & LOOKUP_RCU))
2080 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2082 if (!(nd->flags & LOOKUP_ROOT))
2083 nd->root.mnt = NULL;
2085 return ERR_PTR(-ECHILD);
2088 static const char *trailing_symlink(struct nameidata *nd)
2091 int error = may_follow_link(nd);
2092 if (unlikely(error))
2093 return ERR_PTR(error);
2094 nd->flags |= LOOKUP_PARENT;
2095 nd->stack[0].name = NULL;
2100 static inline int lookup_last(struct nameidata *nd)
2102 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2103 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2105 nd->flags &= ~LOOKUP_PARENT;
2106 return walk_component(nd,
2107 nd->flags & LOOKUP_FOLLOW
2109 ? WALK_PUT | WALK_GET
2114 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2115 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2117 const char *s = path_init(nd, flags);
2122 while (!(err = link_path_walk(s, nd))
2123 && ((err = lookup_last(nd)) > 0)) {
2124 s = trailing_symlink(nd);
2131 err = complete_walk(nd);
2133 if (!err && nd->flags & LOOKUP_DIRECTORY)
2134 if (!d_can_lookup(nd->path.dentry))
2138 nd->path.mnt = NULL;
2139 nd->path.dentry = NULL;
2145 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2146 struct path *path, struct path *root)
2149 struct nameidata nd;
2151 return PTR_ERR(name);
2152 if (unlikely(root)) {
2154 flags |= LOOKUP_ROOT;
2156 set_nameidata(&nd, dfd, name);
2157 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2158 if (unlikely(retval == -ECHILD))
2159 retval = path_lookupat(&nd, flags, path);
2160 if (unlikely(retval == -ESTALE))
2161 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2163 if (likely(!retval))
2164 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2165 restore_nameidata();
2170 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2171 static int path_parentat(struct nameidata *nd, unsigned flags,
2172 struct path *parent)
2174 const char *s = path_init(nd, flags);
2178 err = link_path_walk(s, nd);
2180 err = complete_walk(nd);
2183 nd->path.mnt = NULL;
2184 nd->path.dentry = NULL;
2190 static struct filename *filename_parentat(int dfd, struct filename *name,
2191 unsigned int flags, struct path *parent,
2192 struct qstr *last, int *type)
2195 struct nameidata nd;
2199 set_nameidata(&nd, dfd, name);
2200 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2201 if (unlikely(retval == -ECHILD))
2202 retval = path_parentat(&nd, flags, parent);
2203 if (unlikely(retval == -ESTALE))
2204 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2205 if (likely(!retval)) {
2207 *type = nd.last_type;
2208 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2211 name = ERR_PTR(retval);
2213 restore_nameidata();
2217 /* does lookup, returns the object with parent locked */
2218 struct dentry *kern_path_locked(const char *name, struct path *path)
2220 struct filename *filename;
2225 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2227 if (IS_ERR(filename))
2228 return ERR_CAST(filename);
2229 if (unlikely(type != LAST_NORM)) {
2232 return ERR_PTR(-EINVAL);
2234 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2235 d = __lookup_hash(&last, path->dentry, 0);
2237 mutex_unlock(&path->dentry->d_inode->i_mutex);
2244 int kern_path(const char *name, unsigned int flags, struct path *path)
2246 return filename_lookup(AT_FDCWD, getname_kernel(name),
2249 EXPORT_SYMBOL(kern_path);
2252 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2253 * @dentry: pointer to dentry of the base directory
2254 * @mnt: pointer to vfs mount of the base directory
2255 * @name: pointer to file name
2256 * @flags: lookup flags
2257 * @path: pointer to struct path to fill
2259 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2260 const char *name, unsigned int flags,
2263 struct path root = {.mnt = mnt, .dentry = dentry};
2264 /* the first argument of filename_lookup() is ignored with root */
2265 return filename_lookup(AT_FDCWD, getname_kernel(name),
2266 flags , path, &root);
2268 EXPORT_SYMBOL(vfs_path_lookup);
2271 * lookup_one_len - filesystem helper to lookup single pathname component
2272 * @name: pathname component to lookup
2273 * @base: base directory to lookup from
2274 * @len: maximum length @len should be interpreted to
2276 * Note that this routine is purely a helper for filesystem usage and should
2277 * not be called by generic code.
2279 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2285 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2289 this.hash = full_name_hash(name, len);
2291 return ERR_PTR(-EACCES);
2293 if (unlikely(name[0] == '.')) {
2294 if (len < 2 || (len == 2 && name[1] == '.'))
2295 return ERR_PTR(-EACCES);
2299 c = *(const unsigned char *)name++;
2300 if (c == '/' || c == '\0')
2301 return ERR_PTR(-EACCES);
2304 * See if the low-level filesystem might want
2305 * to use its own hash..
2307 if (base->d_flags & DCACHE_OP_HASH) {
2308 int err = base->d_op->d_hash(base, &this);
2310 return ERR_PTR(err);
2313 err = inode_permission(base->d_inode, MAY_EXEC);
2315 return ERR_PTR(err);
2317 return __lookup_hash(&this, base, 0);
2319 EXPORT_SYMBOL(lookup_one_len);
2321 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2322 struct path *path, int *empty)
2324 return filename_lookup(dfd, getname_flags(name, flags, empty),
2327 EXPORT_SYMBOL(user_path_at_empty);
2330 * NB: most callers don't do anything directly with the reference to the
2331 * to struct filename, but the nd->last pointer points into the name string
2332 * allocated by getname. So we must hold the reference to it until all
2333 * path-walking is complete.
2335 static inline struct filename *
2336 user_path_parent(int dfd, const char __user *path,
2337 struct path *parent,
2342 /* only LOOKUP_REVAL is allowed in extra flags */
2343 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2344 parent, last, type);
2348 * mountpoint_last - look up last component for umount
2349 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2350 * @path: pointer to container for result
2352 * This is a special lookup_last function just for umount. In this case, we
2353 * need to resolve the path without doing any revalidation.
2355 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2356 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2357 * in almost all cases, this lookup will be served out of the dcache. The only
2358 * cases where it won't are if nd->last refers to a symlink or the path is
2359 * bogus and it doesn't exist.
2362 * -error: if there was an error during lookup. This includes -ENOENT if the
2363 * lookup found a negative dentry. The nd->path reference will also be
2366 * 0: if we successfully resolved nd->path and found it to not to be a
2367 * symlink that needs to be followed. "path" will also be populated.
2368 * The nd->path reference will also be put.
2370 * 1: if we successfully resolved nd->last and found it to be a symlink
2371 * that needs to be followed. "path" will be populated with the path
2372 * to the link, and nd->path will *not* be put.
2375 mountpoint_last(struct nameidata *nd, struct path *path)
2378 struct dentry *dentry;
2379 struct dentry *dir = nd->path.dentry;
2381 /* If we're in rcuwalk, drop out of it to handle last component */
2382 if (nd->flags & LOOKUP_RCU) {
2383 if (unlazy_walk(nd, NULL, 0))
2387 nd->flags &= ~LOOKUP_PARENT;
2389 if (unlikely(nd->last_type != LAST_NORM)) {
2390 error = handle_dots(nd, nd->last_type);
2393 dentry = dget(nd->path.dentry);
2397 mutex_lock(&dir->d_inode->i_mutex);
2398 dentry = d_lookup(dir, &nd->last);
2401 * No cached dentry. Mounted dentries are pinned in the cache,
2402 * so that means that this dentry is probably a symlink or the
2403 * path doesn't actually point to a mounted dentry.
2405 dentry = d_alloc(dir, &nd->last);
2407 mutex_unlock(&dir->d_inode->i_mutex);
2410 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2411 if (IS_ERR(dentry)) {
2412 mutex_unlock(&dir->d_inode->i_mutex);
2413 return PTR_ERR(dentry);
2416 mutex_unlock(&dir->d_inode->i_mutex);
2419 if (d_is_negative(dentry)) {
2425 path->dentry = dentry;
2426 path->mnt = nd->path.mnt;
2427 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2428 d_backing_inode(dentry), 0);
2429 if (unlikely(error))
2437 * path_mountpoint - look up a path to be umounted
2438 * @nd: lookup context
2439 * @flags: lookup flags
2440 * @path: pointer to container for result
2442 * Look up the given name, but don't attempt to revalidate the last component.
2443 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2446 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2448 const char *s = path_init(nd, flags);
2452 while (!(err = link_path_walk(s, nd)) &&
2453 (err = mountpoint_last(nd, path)) > 0) {
2454 s = trailing_symlink(nd);
2465 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2468 struct nameidata nd;
2471 return PTR_ERR(name);
2472 set_nameidata(&nd, dfd, name);
2473 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2474 if (unlikely(error == -ECHILD))
2475 error = path_mountpoint(&nd, flags, path);
2476 if (unlikely(error == -ESTALE))
2477 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2479 audit_inode(name, path->dentry, 0);
2480 restore_nameidata();
2486 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2487 * @dfd: directory file descriptor
2488 * @name: pathname from userland
2489 * @flags: lookup flags
2490 * @path: pointer to container to hold result
2492 * A umount is a special case for path walking. We're not actually interested
2493 * in the inode in this situation, and ESTALE errors can be a problem. We
2494 * simply want track down the dentry and vfsmount attached at the mountpoint
2495 * and avoid revalidating the last component.
2497 * Returns 0 and populates "path" on success.
2500 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2503 return filename_mountpoint(dfd, getname(name), path, flags);
2507 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2510 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2512 EXPORT_SYMBOL(kern_path_mountpoint);
2514 int __check_sticky(struct inode *dir, struct inode *inode)
2516 kuid_t fsuid = current_fsuid();
2518 if (uid_eq(inode->i_uid, fsuid))
2520 if (uid_eq(dir->i_uid, fsuid))
2522 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2524 EXPORT_SYMBOL(__check_sticky);
2527 * Check whether we can remove a link victim from directory dir, check
2528 * whether the type of victim is right.
2529 * 1. We can't do it if dir is read-only (done in permission())
2530 * 2. We should have write and exec permissions on dir
2531 * 3. We can't remove anything from append-only dir
2532 * 4. We can't do anything with immutable dir (done in permission())
2533 * 5. If the sticky bit on dir is set we should either
2534 * a. be owner of dir, or
2535 * b. be owner of victim, or
2536 * c. have CAP_FOWNER capability
2537 * 6. If the victim is append-only or immutable we can't do antyhing with
2538 * links pointing to it.
2539 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2540 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2541 * 9. We can't remove a root or mountpoint.
2542 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2543 * nfs_async_unlink().
2545 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2547 struct inode *inode = d_backing_inode(victim);
2550 if (d_is_negative(victim))
2554 BUG_ON(victim->d_parent->d_inode != dir);
2555 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2557 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2563 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2564 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2567 if (!d_is_dir(victim))
2569 if (IS_ROOT(victim))
2571 } else if (d_is_dir(victim))
2573 if (IS_DEADDIR(dir))
2575 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2580 /* Check whether we can create an object with dentry child in directory
2582 * 1. We can't do it if child already exists (open has special treatment for
2583 * this case, but since we are inlined it's OK)
2584 * 2. We can't do it if dir is read-only (done in permission())
2585 * 3. We should have write and exec permissions on dir
2586 * 4. We can't do it if dir is immutable (done in permission())
2588 static inline int may_create(struct inode *dir, struct dentry *child)
2590 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2593 if (IS_DEADDIR(dir))
2595 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2599 * p1 and p2 should be directories on the same fs.
2601 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2606 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2610 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2612 p = d_ancestor(p2, p1);
2614 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2615 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2619 p = d_ancestor(p1, p2);
2621 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2622 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2626 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2627 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2630 EXPORT_SYMBOL(lock_rename);
2632 void unlock_rename(struct dentry *p1, struct dentry *p2)
2634 mutex_unlock(&p1->d_inode->i_mutex);
2636 mutex_unlock(&p2->d_inode->i_mutex);
2637 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2640 EXPORT_SYMBOL(unlock_rename);
2642 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2645 int error = may_create(dir, dentry);
2649 if (!dir->i_op->create)
2650 return -EACCES; /* shouldn't it be ENOSYS? */
2653 error = security_inode_create(dir, dentry, mode);
2656 error = dir->i_op->create(dir, dentry, mode, want_excl);
2658 fsnotify_create(dir, dentry);
2661 EXPORT_SYMBOL(vfs_create);
2663 static int may_open(struct path *path, int acc_mode, int flag)
2665 struct dentry *dentry = path->dentry;
2666 struct inode *inode = dentry->d_inode;
2676 switch (inode->i_mode & S_IFMT) {
2680 if (acc_mode & MAY_WRITE)
2685 if (path->mnt->mnt_flags & MNT_NODEV)
2694 error = inode_permission(inode, acc_mode);
2699 * An append-only file must be opened in append mode for writing.
2701 if (IS_APPEND(inode)) {
2702 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2708 /* O_NOATIME can only be set by the owner or superuser */
2709 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2715 static int handle_truncate(struct file *filp)
2717 struct path *path = &filp->f_path;
2718 struct inode *inode = path->dentry->d_inode;
2719 int error = get_write_access(inode);
2723 * Refuse to truncate files with mandatory locks held on them.
2725 error = locks_verify_locked(filp);
2727 error = security_path_truncate(path);
2729 error = do_truncate(path->dentry, 0,
2730 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2733 put_write_access(inode);
2737 static inline int open_to_namei_flags(int flag)
2739 if ((flag & O_ACCMODE) == 3)
2744 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2746 int error = security_path_mknod(dir, dentry, mode, 0);
2750 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2754 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2758 * Attempt to atomically look up, create and open a file from a negative
2761 * Returns 0 if successful. The file will have been created and attached to
2762 * @file by the filesystem calling finish_open().
2764 * Returns 1 if the file was looked up only or didn't need creating. The
2765 * caller will need to perform the open themselves. @path will have been
2766 * updated to point to the new dentry. This may be negative.
2768 * Returns an error code otherwise.
2770 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2771 struct path *path, struct file *file,
2772 const struct open_flags *op,
2773 bool got_write, bool need_lookup,
2776 struct inode *dir = nd->path.dentry->d_inode;
2777 unsigned open_flag = open_to_namei_flags(op->open_flag);
2781 int create_error = 0;
2782 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2785 BUG_ON(dentry->d_inode);
2787 /* Don't create child dentry for a dead directory. */
2788 if (unlikely(IS_DEADDIR(dir))) {
2794 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2795 mode &= ~current_umask();
2797 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2799 open_flag &= ~O_TRUNC;
2802 * Checking write permission is tricky, bacuse we don't know if we are
2803 * going to actually need it: O_CREAT opens should work as long as the
2804 * file exists. But checking existence breaks atomicity. The trick is
2805 * to check access and if not granted clear O_CREAT from the flags.
2807 * Another problem is returing the "right" error value (e.g. for an
2808 * O_EXCL open we want to return EEXIST not EROFS).
2810 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2811 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2812 if (!(open_flag & O_CREAT)) {
2814 * No O_CREATE -> atomicity not a requirement -> fall
2815 * back to lookup + open
2818 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2819 /* Fall back and fail with the right error */
2820 create_error = -EROFS;
2823 /* No side effects, safe to clear O_CREAT */
2824 create_error = -EROFS;
2825 open_flag &= ~O_CREAT;
2829 if (open_flag & O_CREAT) {
2830 error = may_o_create(&nd->path, dentry, mode);
2832 create_error = error;
2833 if (open_flag & O_EXCL)
2835 open_flag &= ~O_CREAT;
2839 if (nd->flags & LOOKUP_DIRECTORY)
2840 open_flag |= O_DIRECTORY;
2842 file->f_path.dentry = DENTRY_NOT_SET;
2843 file->f_path.mnt = nd->path.mnt;
2844 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2847 if (create_error && error == -ENOENT)
2848 error = create_error;
2852 if (error) { /* returned 1, that is */
2853 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2857 if (file->f_path.dentry) {
2859 dentry = file->f_path.dentry;
2861 if (*opened & FILE_CREATED)
2862 fsnotify_create(dir, dentry);
2863 if (!dentry->d_inode) {
2864 WARN_ON(*opened & FILE_CREATED);
2866 error = create_error;
2870 if (excl && !(*opened & FILE_CREATED)) {
2879 * We didn't have the inode before the open, so check open permission
2882 acc_mode = op->acc_mode;
2883 if (*opened & FILE_CREATED) {
2884 WARN_ON(!(open_flag & O_CREAT));
2885 fsnotify_create(dir, dentry);
2886 acc_mode = MAY_OPEN;
2888 error = may_open(&file->f_path, acc_mode, open_flag);
2898 dentry = lookup_real(dir, dentry, nd->flags);
2900 return PTR_ERR(dentry);
2903 int open_flag = op->open_flag;
2905 error = create_error;
2906 if ((open_flag & O_EXCL)) {
2907 if (!dentry->d_inode)
2909 } else if (!dentry->d_inode) {
2911 } else if ((open_flag & O_TRUNC) &&
2915 /* will fail later, go on to get the right error */
2919 path->dentry = dentry;
2920 path->mnt = nd->path.mnt;
2925 * Look up and maybe create and open the last component.
2927 * Must be called with i_mutex held on parent.
2929 * Returns 0 if the file was successfully atomically created (if necessary) and
2930 * opened. In this case the file will be returned attached to @file.
2932 * Returns 1 if the file was not completely opened at this time, though lookups
2933 * and creations will have been performed and the dentry returned in @path will
2934 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2935 * specified then a negative dentry may be returned.
2937 * An error code is returned otherwise.
2939 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2940 * cleared otherwise prior to returning.
2942 static int lookup_open(struct nameidata *nd, struct path *path,
2944 const struct open_flags *op,
2945 bool got_write, int *opened)
2947 struct dentry *dir = nd->path.dentry;
2948 struct inode *dir_inode = dir->d_inode;
2949 struct dentry *dentry;
2953 *opened &= ~FILE_CREATED;
2954 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2956 return PTR_ERR(dentry);
2958 /* Cached positive dentry: will open in f_op->open */
2959 if (!need_lookup && dentry->d_inode)
2962 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2963 return atomic_open(nd, dentry, path, file, op, got_write,
2964 need_lookup, opened);
2968 BUG_ON(dentry->d_inode);
2970 dentry = lookup_real(dir_inode, dentry, nd->flags);
2972 return PTR_ERR(dentry);
2975 /* Negative dentry, just create the file */
2976 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2977 umode_t mode = op->mode;
2978 if (!IS_POSIXACL(dir->d_inode))
2979 mode &= ~current_umask();
2981 * This write is needed to ensure that a
2982 * rw->ro transition does not occur between
2983 * the time when the file is created and when
2984 * a permanent write count is taken through
2985 * the 'struct file' in finish_open().
2991 *opened |= FILE_CREATED;
2992 error = security_path_mknod(&nd->path, dentry, mode, 0);
2995 error = vfs_create(dir->d_inode, dentry, mode,
2996 nd->flags & LOOKUP_EXCL);
3001 path->dentry = dentry;
3002 path->mnt = nd->path.mnt;
3011 * Handle the last step of open()
3013 static int do_last(struct nameidata *nd,
3014 struct file *file, const struct open_flags *op,
3017 struct dentry *dir = nd->path.dentry;
3018 int open_flag = op->open_flag;
3019 bool will_truncate = (open_flag & O_TRUNC) != 0;
3020 bool got_write = false;
3021 int acc_mode = op->acc_mode;
3023 struct inode *inode;
3024 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3026 bool retried = false;
3029 nd->flags &= ~LOOKUP_PARENT;
3030 nd->flags |= op->intent;
3032 if (nd->last_type != LAST_NORM) {
3033 error = handle_dots(nd, nd->last_type);
3034 if (unlikely(error))
3039 if (!(open_flag & O_CREAT)) {
3040 if (nd->last.name[nd->last.len])
3041 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3042 /* we _can_ be in RCU mode here */
3043 error = lookup_fast(nd, &path, &inode, &seq);
3050 BUG_ON(nd->inode != dir->d_inode);
3052 /* create side of things */
3054 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3055 * has been cleared when we got to the last component we are
3058 error = complete_walk(nd);
3062 audit_inode(nd->name, dir, LOOKUP_PARENT);
3063 /* trailing slashes? */
3064 if (unlikely(nd->last.name[nd->last.len]))
3069 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3070 error = mnt_want_write(nd->path.mnt);
3074 * do _not_ fail yet - we might not need that or fail with
3075 * a different error; let lookup_open() decide; we'll be
3076 * dropping this one anyway.
3079 mutex_lock(&dir->d_inode->i_mutex);
3080 error = lookup_open(nd, &path, file, op, got_write, opened);
3081 mutex_unlock(&dir->d_inode->i_mutex);
3087 if ((*opened & FILE_CREATED) ||
3088 !S_ISREG(file_inode(file)->i_mode))
3089 will_truncate = false;
3091 audit_inode(nd->name, file->f_path.dentry, 0);
3095 if (*opened & FILE_CREATED) {
3096 /* Don't check for write permission, don't truncate */
3097 open_flag &= ~O_TRUNC;
3098 will_truncate = false;
3099 acc_mode = MAY_OPEN;
3100 path_to_nameidata(&path, nd);
3101 goto finish_open_created;
3105 * create/update audit record if it already exists.
3107 if (d_is_positive(path.dentry))
3108 audit_inode(nd->name, path.dentry, 0);
3111 * If atomic_open() acquired write access it is dropped now due to
3112 * possible mount and symlink following (this might be optimized away if
3116 mnt_drop_write(nd->path.mnt);
3120 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3121 path_to_nameidata(&path, nd);
3125 error = follow_managed(&path, nd);
3126 if (unlikely(error < 0))
3129 BUG_ON(nd->flags & LOOKUP_RCU);
3130 inode = d_backing_inode(path.dentry);
3131 seq = 0; /* out of RCU mode, so the value doesn't matter */
3132 if (unlikely(d_is_negative(path.dentry))) {
3133 path_to_nameidata(&path, nd);
3139 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3141 if (unlikely(error))
3144 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3145 path_to_nameidata(&path, nd);
3149 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3150 path_to_nameidata(&path, nd);
3152 save_parent.dentry = nd->path.dentry;
3153 save_parent.mnt = mntget(path.mnt);
3154 nd->path.dentry = path.dentry;
3159 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3161 error = complete_walk(nd);
3163 path_put(&save_parent);
3166 audit_inode(nd->name, nd->path.dentry, 0);
3168 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3171 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3173 if (!d_is_reg(nd->path.dentry))
3174 will_truncate = false;
3176 if (will_truncate) {
3177 error = mnt_want_write(nd->path.mnt);
3182 finish_open_created:
3183 error = may_open(&nd->path, acc_mode, open_flag);
3187 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3188 error = vfs_open(&nd->path, file, current_cred());
3190 *opened |= FILE_OPENED;
3192 if (error == -EOPENSTALE)
3197 error = open_check_o_direct(file);
3200 error = ima_file_check(file, op->acc_mode, *opened);
3204 if (will_truncate) {
3205 error = handle_truncate(file);
3211 mnt_drop_write(nd->path.mnt);
3212 path_put(&save_parent);
3220 /* If no saved parent or already retried then can't retry */
3221 if (!save_parent.dentry || retried)
3224 BUG_ON(save_parent.dentry != dir);
3225 path_put(&nd->path);
3226 nd->path = save_parent;
3227 nd->inode = dir->d_inode;
3228 save_parent.mnt = NULL;
3229 save_parent.dentry = NULL;
3231 mnt_drop_write(nd->path.mnt);
3238 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3239 const struct open_flags *op,
3240 struct file *file, int *opened)
3242 static const struct qstr name = QSTR_INIT("/", 1);
3243 struct dentry *child;
3246 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3247 if (unlikely(error))
3249 error = mnt_want_write(path.mnt);
3250 if (unlikely(error))
3252 dir = path.dentry->d_inode;
3253 /* we want directory to be writable */
3254 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3257 if (!dir->i_op->tmpfile) {
3258 error = -EOPNOTSUPP;
3261 child = d_alloc(path.dentry, &name);
3262 if (unlikely(!child)) {
3267 path.dentry = child;
3268 error = dir->i_op->tmpfile(dir, child, op->mode);
3271 audit_inode(nd->name, child, 0);
3272 /* Don't check for other permissions, the inode was just created */
3273 error = may_open(&path, MAY_OPEN, op->open_flag);
3276 file->f_path.mnt = path.mnt;
3277 error = finish_open(file, child, NULL, opened);
3280 error = open_check_o_direct(file);
3283 } else if (!(op->open_flag & O_EXCL)) {
3284 struct inode *inode = file_inode(file);
3285 spin_lock(&inode->i_lock);
3286 inode->i_state |= I_LINKABLE;
3287 spin_unlock(&inode->i_lock);
3290 mnt_drop_write(path.mnt);
3296 static struct file *path_openat(struct nameidata *nd,
3297 const struct open_flags *op, unsigned flags)
3304 file = get_empty_filp();
3308 file->f_flags = op->open_flag;
3310 if (unlikely(file->f_flags & __O_TMPFILE)) {
3311 error = do_tmpfile(nd, flags, op, file, &opened);
3315 s = path_init(nd, flags);
3320 while (!(error = link_path_walk(s, nd)) &&
3321 (error = do_last(nd, file, op, &opened)) > 0) {
3322 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3323 s = trailing_symlink(nd);
3331 if (!(opened & FILE_OPENED)) {
3335 if (unlikely(error)) {
3336 if (error == -EOPENSTALE) {
3337 if (flags & LOOKUP_RCU)
3342 file = ERR_PTR(error);
3347 struct file *do_filp_open(int dfd, struct filename *pathname,
3348 const struct open_flags *op)
3350 struct nameidata nd;
3351 int flags = op->lookup_flags;
3354 set_nameidata(&nd, dfd, pathname);
3355 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3356 if (unlikely(filp == ERR_PTR(-ECHILD)))
3357 filp = path_openat(&nd, op, flags);
3358 if (unlikely(filp == ERR_PTR(-ESTALE)))
3359 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3360 restore_nameidata();
3364 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3365 const char *name, const struct open_flags *op)
3367 struct nameidata nd;
3369 struct filename *filename;
3370 int flags = op->lookup_flags | LOOKUP_ROOT;
3373 nd.root.dentry = dentry;
3375 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3376 return ERR_PTR(-ELOOP);
3378 filename = getname_kernel(name);
3379 if (unlikely(IS_ERR(filename)))
3380 return ERR_CAST(filename);
3382 set_nameidata(&nd, -1, filename);
3383 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3384 if (unlikely(file == ERR_PTR(-ECHILD)))
3385 file = path_openat(&nd, op, flags);
3386 if (unlikely(file == ERR_PTR(-ESTALE)))
3387 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3388 restore_nameidata();
3393 static struct dentry *filename_create(int dfd, struct filename *name,
3394 struct path *path, unsigned int lookup_flags)
3396 struct dentry *dentry = ERR_PTR(-EEXIST);
3401 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3404 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3405 * other flags passed in are ignored!
3407 lookup_flags &= LOOKUP_REVAL;
3409 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3411 return ERR_CAST(name);
3414 * Yucky last component or no last component at all?
3415 * (foo/., foo/.., /////)
3417 if (unlikely(type != LAST_NORM))
3420 /* don't fail immediately if it's r/o, at least try to report other errors */
3421 err2 = mnt_want_write(path->mnt);
3423 * Do the final lookup.
3425 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3426 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3427 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3432 if (d_is_positive(dentry))
3436 * Special case - lookup gave negative, but... we had foo/bar/
3437 * From the vfs_mknod() POV we just have a negative dentry -
3438 * all is fine. Let's be bastards - you had / on the end, you've
3439 * been asking for (non-existent) directory. -ENOENT for you.
3441 if (unlikely(!is_dir && last.name[last.len])) {
3445 if (unlikely(err2)) {
3453 dentry = ERR_PTR(error);
3455 mutex_unlock(&path->dentry->d_inode->i_mutex);
3457 mnt_drop_write(path->mnt);
3464 struct dentry *kern_path_create(int dfd, const char *pathname,
3465 struct path *path, unsigned int lookup_flags)
3467 return filename_create(dfd, getname_kernel(pathname),
3468 path, lookup_flags);
3470 EXPORT_SYMBOL(kern_path_create);
3472 void done_path_create(struct path *path, struct dentry *dentry)
3475 mutex_unlock(&path->dentry->d_inode->i_mutex);
3476 mnt_drop_write(path->mnt);
3479 EXPORT_SYMBOL(done_path_create);
3481 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3482 struct path *path, unsigned int lookup_flags)
3484 return filename_create(dfd, getname(pathname), path, lookup_flags);
3486 EXPORT_SYMBOL(user_path_create);
3488 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3490 int error = may_create(dir, dentry);
3495 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3498 if (!dir->i_op->mknod)
3501 error = devcgroup_inode_mknod(mode, dev);
3505 error = security_inode_mknod(dir, dentry, mode, dev);
3509 error = dir->i_op->mknod(dir, dentry, mode, dev);
3511 fsnotify_create(dir, dentry);
3514 EXPORT_SYMBOL(vfs_mknod);
3516 static int may_mknod(umode_t mode)
3518 switch (mode & S_IFMT) {
3524 case 0: /* zero mode translates to S_IFREG */
3533 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3536 struct dentry *dentry;
3539 unsigned int lookup_flags = 0;
3541 error = may_mknod(mode);
3545 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3547 return PTR_ERR(dentry);
3549 if (!IS_POSIXACL(path.dentry->d_inode))
3550 mode &= ~current_umask();
3551 error = security_path_mknod(&path, dentry, mode, dev);
3554 switch (mode & S_IFMT) {
3555 case 0: case S_IFREG:
3556 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3558 case S_IFCHR: case S_IFBLK:
3559 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3560 new_decode_dev(dev));
3562 case S_IFIFO: case S_IFSOCK:
3563 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3567 done_path_create(&path, dentry);
3568 if (retry_estale(error, lookup_flags)) {
3569 lookup_flags |= LOOKUP_REVAL;
3575 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3577 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3580 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3582 int error = may_create(dir, dentry);
3583 unsigned max_links = dir->i_sb->s_max_links;
3588 if (!dir->i_op->mkdir)
3591 mode &= (S_IRWXUGO|S_ISVTX);
3592 error = security_inode_mkdir(dir, dentry, mode);
3596 if (max_links && dir->i_nlink >= max_links)
3599 error = dir->i_op->mkdir(dir, dentry, mode);
3601 fsnotify_mkdir(dir, dentry);
3604 EXPORT_SYMBOL(vfs_mkdir);
3606 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3608 struct dentry *dentry;
3611 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3614 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3616 return PTR_ERR(dentry);
3618 if (!IS_POSIXACL(path.dentry->d_inode))
3619 mode &= ~current_umask();
3620 error = security_path_mkdir(&path, dentry, mode);
3622 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3623 done_path_create(&path, dentry);
3624 if (retry_estale(error, lookup_flags)) {
3625 lookup_flags |= LOOKUP_REVAL;
3631 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3633 return sys_mkdirat(AT_FDCWD, pathname, mode);
3637 * The dentry_unhash() helper will try to drop the dentry early: we
3638 * should have a usage count of 1 if we're the only user of this
3639 * dentry, and if that is true (possibly after pruning the dcache),
3640 * then we drop the dentry now.
3642 * A low-level filesystem can, if it choses, legally
3645 * if (!d_unhashed(dentry))
3648 * if it cannot handle the case of removing a directory
3649 * that is still in use by something else..
3651 void dentry_unhash(struct dentry *dentry)
3653 shrink_dcache_parent(dentry);
3654 spin_lock(&dentry->d_lock);
3655 if (dentry->d_lockref.count == 1)
3657 spin_unlock(&dentry->d_lock);
3659 EXPORT_SYMBOL(dentry_unhash);
3661 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3663 int error = may_delete(dir, dentry, 1);
3668 if (!dir->i_op->rmdir)
3672 mutex_lock(&dentry->d_inode->i_mutex);
3675 if (is_local_mountpoint(dentry))
3678 error = security_inode_rmdir(dir, dentry);
3682 shrink_dcache_parent(dentry);
3683 error = dir->i_op->rmdir(dir, dentry);
3687 dentry->d_inode->i_flags |= S_DEAD;
3689 detach_mounts(dentry);
3692 mutex_unlock(&dentry->d_inode->i_mutex);
3698 EXPORT_SYMBOL(vfs_rmdir);
3700 static long do_rmdir(int dfd, const char __user *pathname)
3703 struct filename *name;
3704 struct dentry *dentry;
3708 unsigned int lookup_flags = 0;
3710 name = user_path_parent(dfd, pathname,
3711 &path, &last, &type, lookup_flags);
3713 return PTR_ERR(name);
3727 error = mnt_want_write(path.mnt);
3731 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3732 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3733 error = PTR_ERR(dentry);
3736 if (!dentry->d_inode) {
3740 error = security_path_rmdir(&path, dentry);
3743 error = vfs_rmdir(path.dentry->d_inode, dentry);
3747 mutex_unlock(&path.dentry->d_inode->i_mutex);
3748 mnt_drop_write(path.mnt);
3752 if (retry_estale(error, lookup_flags)) {
3753 lookup_flags |= LOOKUP_REVAL;
3759 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3761 return do_rmdir(AT_FDCWD, pathname);
3765 * vfs_unlink - unlink a filesystem object
3766 * @dir: parent directory
3768 * @delegated_inode: returns victim inode, if the inode is delegated.
3770 * The caller must hold dir->i_mutex.
3772 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3773 * return a reference to the inode in delegated_inode. The caller
3774 * should then break the delegation on that inode and retry. Because
3775 * breaking a delegation may take a long time, the caller should drop
3776 * dir->i_mutex before doing so.
3778 * Alternatively, a caller may pass NULL for delegated_inode. This may
3779 * be appropriate for callers that expect the underlying filesystem not
3780 * to be NFS exported.
3782 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3784 struct inode *target = dentry->d_inode;
3785 int error = may_delete(dir, dentry, 0);
3790 if (!dir->i_op->unlink)
3793 mutex_lock(&target->i_mutex);
3794 if (is_local_mountpoint(dentry))
3797 error = security_inode_unlink(dir, dentry);
3799 error = try_break_deleg(target, delegated_inode);
3802 error = dir->i_op->unlink(dir, dentry);
3805 detach_mounts(dentry);
3810 mutex_unlock(&target->i_mutex);
3812 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3813 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3814 fsnotify_link_count(target);
3820 EXPORT_SYMBOL(vfs_unlink);
3823 * Make sure that the actual truncation of the file will occur outside its
3824 * directory's i_mutex. Truncate can take a long time if there is a lot of
3825 * writeout happening, and we don't want to prevent access to the directory
3826 * while waiting on the I/O.
3828 static long do_unlinkat(int dfd, const char __user *pathname)
3831 struct filename *name;
3832 struct dentry *dentry;
3836 struct inode *inode = NULL;
3837 struct inode *delegated_inode = NULL;
3838 unsigned int lookup_flags = 0;
3840 name = user_path_parent(dfd, pathname,
3841 &path, &last, &type, lookup_flags);
3843 return PTR_ERR(name);
3846 if (type != LAST_NORM)
3849 error = mnt_want_write(path.mnt);
3853 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3854 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3855 error = PTR_ERR(dentry);
3856 if (!IS_ERR(dentry)) {
3857 /* Why not before? Because we want correct error value */
3858 if (last.name[last.len])
3860 inode = dentry->d_inode;
3861 if (d_is_negative(dentry))
3864 error = security_path_unlink(&path, dentry);
3867 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3871 mutex_unlock(&path.dentry->d_inode->i_mutex);
3873 iput(inode); /* truncate the inode here */
3875 if (delegated_inode) {
3876 error = break_deleg_wait(&delegated_inode);
3880 mnt_drop_write(path.mnt);
3884 if (retry_estale(error, lookup_flags)) {
3885 lookup_flags |= LOOKUP_REVAL;
3892 if (d_is_negative(dentry))
3894 else if (d_is_dir(dentry))
3901 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3903 if ((flag & ~AT_REMOVEDIR) != 0)
3906 if (flag & AT_REMOVEDIR)
3907 return do_rmdir(dfd, pathname);
3909 return do_unlinkat(dfd, pathname);
3912 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3914 return do_unlinkat(AT_FDCWD, pathname);
3917 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3919 int error = may_create(dir, dentry);
3924 if (!dir->i_op->symlink)
3927 error = security_inode_symlink(dir, dentry, oldname);
3931 error = dir->i_op->symlink(dir, dentry, oldname);
3933 fsnotify_create(dir, dentry);
3936 EXPORT_SYMBOL(vfs_symlink);
3938 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3939 int, newdfd, const char __user *, newname)
3942 struct filename *from;
3943 struct dentry *dentry;
3945 unsigned int lookup_flags = 0;
3947 from = getname(oldname);
3949 return PTR_ERR(from);
3951 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3952 error = PTR_ERR(dentry);
3956 error = security_path_symlink(&path, dentry, from->name);
3958 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3959 done_path_create(&path, dentry);
3960 if (retry_estale(error, lookup_flags)) {
3961 lookup_flags |= LOOKUP_REVAL;
3969 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3971 return sys_symlinkat(oldname, AT_FDCWD, newname);
3975 * vfs_link - create a new link
3976 * @old_dentry: object to be linked
3978 * @new_dentry: where to create the new link
3979 * @delegated_inode: returns inode needing a delegation break
3981 * The caller must hold dir->i_mutex
3983 * If vfs_link discovers a delegation on the to-be-linked file in need
3984 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3985 * inode in delegated_inode. The caller should then break the delegation
3986 * and retry. Because breaking a delegation may take a long time, the
3987 * caller should drop the i_mutex before doing so.
3989 * Alternatively, a caller may pass NULL for delegated_inode. This may
3990 * be appropriate for callers that expect the underlying filesystem not
3991 * to be NFS exported.
3993 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3995 struct inode *inode = old_dentry->d_inode;
3996 unsigned max_links = dir->i_sb->s_max_links;
4002 error = may_create(dir, new_dentry);
4006 if (dir->i_sb != inode->i_sb)
4010 * A link to an append-only or immutable file cannot be created.
4012 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4014 if (!dir->i_op->link)
4016 if (S_ISDIR(inode->i_mode))
4019 error = security_inode_link(old_dentry, dir, new_dentry);
4023 mutex_lock(&inode->i_mutex);
4024 /* Make sure we don't allow creating hardlink to an unlinked file */
4025 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4027 else if (max_links && inode->i_nlink >= max_links)
4030 error = try_break_deleg(inode, delegated_inode);
4032 error = dir->i_op->link(old_dentry, dir, new_dentry);
4035 if (!error && (inode->i_state & I_LINKABLE)) {
4036 spin_lock(&inode->i_lock);
4037 inode->i_state &= ~I_LINKABLE;
4038 spin_unlock(&inode->i_lock);
4040 mutex_unlock(&inode->i_mutex);
4042 fsnotify_link(dir, inode, new_dentry);
4045 EXPORT_SYMBOL(vfs_link);
4048 * Hardlinks are often used in delicate situations. We avoid
4049 * security-related surprises by not following symlinks on the
4052 * We don't follow them on the oldname either to be compatible
4053 * with linux 2.0, and to avoid hard-linking to directories
4054 * and other special files. --ADM
4056 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4057 int, newdfd, const char __user *, newname, int, flags)
4059 struct dentry *new_dentry;
4060 struct path old_path, new_path;
4061 struct inode *delegated_inode = NULL;
4065 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4068 * To use null names we require CAP_DAC_READ_SEARCH
4069 * This ensures that not everyone will be able to create
4070 * handlink using the passed filedescriptor.
4072 if (flags & AT_EMPTY_PATH) {
4073 if (!capable(CAP_DAC_READ_SEARCH))
4078 if (flags & AT_SYMLINK_FOLLOW)
4079 how |= LOOKUP_FOLLOW;
4081 error = user_path_at(olddfd, oldname, how, &old_path);
4085 new_dentry = user_path_create(newdfd, newname, &new_path,
4086 (how & LOOKUP_REVAL));
4087 error = PTR_ERR(new_dentry);
4088 if (IS_ERR(new_dentry))
4092 if (old_path.mnt != new_path.mnt)
4094 error = may_linkat(&old_path);
4095 if (unlikely(error))
4097 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4100 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4102 done_path_create(&new_path, new_dentry);
4103 if (delegated_inode) {
4104 error = break_deleg_wait(&delegated_inode);
4106 path_put(&old_path);
4110 if (retry_estale(error, how)) {
4111 path_put(&old_path);
4112 how |= LOOKUP_REVAL;
4116 path_put(&old_path);
4121 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4123 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4127 * vfs_rename - rename a filesystem object
4128 * @old_dir: parent of source
4129 * @old_dentry: source
4130 * @new_dir: parent of destination
4131 * @new_dentry: destination
4132 * @delegated_inode: returns an inode needing a delegation break
4133 * @flags: rename flags
4135 * The caller must hold multiple mutexes--see lock_rename()).
4137 * If vfs_rename discovers a delegation in need of breaking at either
4138 * the source or destination, it will return -EWOULDBLOCK and return a
4139 * reference to the inode in delegated_inode. The caller should then
4140 * break the delegation and retry. Because breaking a delegation may
4141 * take a long time, the caller should drop all locks before doing
4144 * Alternatively, a caller may pass NULL for delegated_inode. This may
4145 * be appropriate for callers that expect the underlying filesystem not
4146 * to be NFS exported.
4148 * The worst of all namespace operations - renaming directory. "Perverted"
4149 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4151 * a) we can get into loop creation.
4152 * b) race potential - two innocent renames can create a loop together.
4153 * That's where 4.4 screws up. Current fix: serialization on
4154 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4156 * c) we have to lock _four_ objects - parents and victim (if it exists),
4157 * and source (if it is not a directory).
4158 * And that - after we got ->i_mutex on parents (until then we don't know
4159 * whether the target exists). Solution: try to be smart with locking
4160 * order for inodes. We rely on the fact that tree topology may change
4161 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4162 * move will be locked. Thus we can rank directories by the tree
4163 * (ancestors first) and rank all non-directories after them.
4164 * That works since everybody except rename does "lock parent, lookup,
4165 * lock child" and rename is under ->s_vfs_rename_mutex.
4166 * HOWEVER, it relies on the assumption that any object with ->lookup()
4167 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4168 * we'd better make sure that there's no link(2) for them.
4169 * d) conversion from fhandle to dentry may come in the wrong moment - when
4170 * we are removing the target. Solution: we will have to grab ->i_mutex
4171 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4172 * ->i_mutex on parents, which works but leads to some truly excessive
4175 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4176 struct inode *new_dir, struct dentry *new_dentry,
4177 struct inode **delegated_inode, unsigned int flags)
4180 bool is_dir = d_is_dir(old_dentry);
4181 const unsigned char *old_name;
4182 struct inode *source = old_dentry->d_inode;
4183 struct inode *target = new_dentry->d_inode;
4184 bool new_is_dir = false;
4185 unsigned max_links = new_dir->i_sb->s_max_links;
4187 if (source == target)
4190 error = may_delete(old_dir, old_dentry, is_dir);
4195 error = may_create(new_dir, new_dentry);
4197 new_is_dir = d_is_dir(new_dentry);
4199 if (!(flags & RENAME_EXCHANGE))
4200 error = may_delete(new_dir, new_dentry, is_dir);
4202 error = may_delete(new_dir, new_dentry, new_is_dir);
4207 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4210 if (flags && !old_dir->i_op->rename2)
4214 * If we are going to change the parent - check write permissions,
4215 * we'll need to flip '..'.
4217 if (new_dir != old_dir) {
4219 error = inode_permission(source, MAY_WRITE);
4223 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4224 error = inode_permission(target, MAY_WRITE);
4230 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4235 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4237 if (!is_dir || (flags & RENAME_EXCHANGE))
4238 lock_two_nondirectories(source, target);
4240 mutex_lock(&target->i_mutex);
4243 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4246 if (max_links && new_dir != old_dir) {
4248 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4250 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4251 old_dir->i_nlink >= max_links)
4254 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4255 shrink_dcache_parent(new_dentry);
4257 error = try_break_deleg(source, delegated_inode);
4261 if (target && !new_is_dir) {
4262 error = try_break_deleg(target, delegated_inode);
4266 if (!old_dir->i_op->rename2) {
4267 error = old_dir->i_op->rename(old_dir, old_dentry,
4268 new_dir, new_dentry);
4270 WARN_ON(old_dir->i_op->rename != NULL);
4271 error = old_dir->i_op->rename2(old_dir, old_dentry,
4272 new_dir, new_dentry, flags);
4277 if (!(flags & RENAME_EXCHANGE) && target) {
4279 target->i_flags |= S_DEAD;
4280 dont_mount(new_dentry);
4281 detach_mounts(new_dentry);
4283 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4284 if (!(flags & RENAME_EXCHANGE))
4285 d_move(old_dentry, new_dentry);
4287 d_exchange(old_dentry, new_dentry);
4290 if (!is_dir || (flags & RENAME_EXCHANGE))
4291 unlock_two_nondirectories(source, target);
4293 mutex_unlock(&target->i_mutex);
4296 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4297 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4298 if (flags & RENAME_EXCHANGE) {
4299 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4300 new_is_dir, NULL, new_dentry);
4303 fsnotify_oldname_free(old_name);
4307 EXPORT_SYMBOL(vfs_rename);
4309 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4310 int, newdfd, const char __user *, newname, unsigned int, flags)
4312 struct dentry *old_dentry, *new_dentry;
4313 struct dentry *trap;
4314 struct path old_path, new_path;
4315 struct qstr old_last, new_last;
4316 int old_type, new_type;
4317 struct inode *delegated_inode = NULL;
4318 struct filename *from;
4319 struct filename *to;
4320 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4321 bool should_retry = false;
4324 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4327 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4328 (flags & RENAME_EXCHANGE))
4331 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4334 if (flags & RENAME_EXCHANGE)
4338 from = user_path_parent(olddfd, oldname,
4339 &old_path, &old_last, &old_type, lookup_flags);
4341 error = PTR_ERR(from);
4345 to = user_path_parent(newdfd, newname,
4346 &new_path, &new_last, &new_type, lookup_flags);
4348 error = PTR_ERR(to);
4353 if (old_path.mnt != new_path.mnt)
4357 if (old_type != LAST_NORM)
4360 if (flags & RENAME_NOREPLACE)
4362 if (new_type != LAST_NORM)
4365 error = mnt_want_write(old_path.mnt);
4370 trap = lock_rename(new_path.dentry, old_path.dentry);
4372 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4373 error = PTR_ERR(old_dentry);
4374 if (IS_ERR(old_dentry))
4376 /* source must exist */
4378 if (d_is_negative(old_dentry))
4380 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4381 error = PTR_ERR(new_dentry);
4382 if (IS_ERR(new_dentry))
4385 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4387 if (flags & RENAME_EXCHANGE) {
4389 if (d_is_negative(new_dentry))
4392 if (!d_is_dir(new_dentry)) {
4394 if (new_last.name[new_last.len])
4398 /* unless the source is a directory trailing slashes give -ENOTDIR */
4399 if (!d_is_dir(old_dentry)) {
4401 if (old_last.name[old_last.len])
4403 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4406 /* source should not be ancestor of target */
4408 if (old_dentry == trap)
4410 /* target should not be an ancestor of source */
4411 if (!(flags & RENAME_EXCHANGE))
4413 if (new_dentry == trap)
4416 error = security_path_rename(&old_path, old_dentry,
4417 &new_path, new_dentry, flags);
4420 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4421 new_path.dentry->d_inode, new_dentry,
4422 &delegated_inode, flags);
4428 unlock_rename(new_path.dentry, old_path.dentry);
4429 if (delegated_inode) {
4430 error = break_deleg_wait(&delegated_inode);
4434 mnt_drop_write(old_path.mnt);
4436 if (retry_estale(error, lookup_flags))
4437 should_retry = true;
4438 path_put(&new_path);
4441 path_put(&old_path);
4444 should_retry = false;
4445 lookup_flags |= LOOKUP_REVAL;
4452 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4453 int, newdfd, const char __user *, newname)
4455 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4458 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4460 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4463 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4465 int error = may_create(dir, dentry);
4469 if (!dir->i_op->mknod)
4472 return dir->i_op->mknod(dir, dentry,
4473 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4475 EXPORT_SYMBOL(vfs_whiteout);
4477 int readlink_copy(char __user *buffer, int buflen, const char *link)
4479 int len = PTR_ERR(link);
4484 if (len > (unsigned) buflen)
4486 if (copy_to_user(buffer, link, len))
4491 EXPORT_SYMBOL(readlink_copy);
4494 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4495 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4496 * using) it for any given inode is up to filesystem.
4498 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4501 struct inode *inode = d_inode(dentry);
4502 const char *link = inode->i_link;
4506 link = inode->i_op->follow_link(dentry, &cookie);
4508 return PTR_ERR(link);
4510 res = readlink_copy(buffer, buflen, link);
4511 if (inode->i_op->put_link)
4512 inode->i_op->put_link(inode, cookie);
4515 EXPORT_SYMBOL(generic_readlink);
4517 /* get the link contents into pagecache */
4518 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4522 struct address_space *mapping = dentry->d_inode->i_mapping;
4523 page = read_mapping_page(mapping, 0, NULL);
4528 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4532 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4534 struct page *page = NULL;
4535 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4538 page_cache_release(page);
4542 EXPORT_SYMBOL(page_readlink);
4544 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4546 struct page *page = NULL;
4547 char *res = page_getlink(dentry, &page);
4552 EXPORT_SYMBOL(page_follow_link_light);
4554 void page_put_link(struct inode *unused, void *cookie)
4556 struct page *page = cookie;
4558 page_cache_release(page);
4560 EXPORT_SYMBOL(page_put_link);
4563 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4565 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4567 struct address_space *mapping = inode->i_mapping;
4572 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4574 flags |= AOP_FLAG_NOFS;
4577 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4578 flags, &page, &fsdata);
4582 kaddr = kmap_atomic(page);
4583 memcpy(kaddr, symname, len-1);
4584 kunmap_atomic(kaddr);
4586 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4593 mark_inode_dirty(inode);
4598 EXPORT_SYMBOL(__page_symlink);
4600 int page_symlink(struct inode *inode, const char *symname, int len)
4602 return __page_symlink(inode, symname, len,
4603 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4605 EXPORT_SYMBOL(page_symlink);
4607 const struct inode_operations page_symlink_inode_operations = {
4608 .readlink = generic_readlink,
4609 .follow_link = page_follow_link_light,
4610 .put_link = page_put_link,
4612 EXPORT_SYMBOL(page_symlink_inode_operations);