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 */
502 unsigned seq, m_seq, root_seq;
505 int total_link_count;
512 } *stack, internal[EMBEDDED_LEVELS];
515 static struct nameidata *set_nameidata(struct nameidata *p)
517 struct nameidata *old = current->nameidata;
518 p->stack = p->internal;
519 p->total_link_count = old ? old->total_link_count : 0;
520 current->nameidata = p;
524 static void restore_nameidata(struct nameidata *old)
526 struct nameidata *now = current->nameidata;
528 current->nameidata = old;
530 old->total_link_count = now->total_link_count;
531 if (now->stack != now->internal) {
533 now->stack = now->internal;
537 static int __nd_alloc_stack(struct nameidata *nd)
541 if (nd->flags & LOOKUP_RCU) {
542 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
547 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
552 memcpy(p, nd->internal, sizeof(nd->internal));
557 static inline int nd_alloc_stack(struct nameidata *nd)
559 if (likely(nd->depth != EMBEDDED_LEVELS))
561 if (likely(nd->stack != nd->internal))
563 return __nd_alloc_stack(nd);
566 static void drop_links(struct nameidata *nd)
570 struct saved *last = nd->stack + i;
571 struct inode *inode = last->inode;
572 if (last->cookie && inode->i_op->put_link) {
573 inode->i_op->put_link(inode, last->cookie);
579 static void terminate_walk(struct nameidata *nd)
582 if (!(nd->flags & LOOKUP_RCU)) {
585 for (i = 0; i < nd->depth; i++)
586 path_put(&nd->stack[i].link);
588 nd->flags &= ~LOOKUP_RCU;
589 if (!(nd->flags & LOOKUP_ROOT))
596 /* path_put is needed afterwards regardless of success or failure */
597 static bool legitimize_path(struct nameidata *nd,
598 struct path *path, unsigned seq)
600 int res = __legitimize_mnt(path->mnt, nd->m_seq);
607 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
611 return !read_seqcount_retry(&path->dentry->d_seq, seq);
614 static bool legitimize_links(struct nameidata *nd)
617 for (i = 0; i < nd->depth; i++) {
618 struct saved *last = nd->stack + i;
619 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
629 * Path walking has 2 modes, rcu-walk and ref-walk (see
630 * Documentation/filesystems/path-lookup.txt). In situations when we can't
631 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
632 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
633 * mode. Refcounts are grabbed at the last known good point before rcu-walk
634 * got stuck, so ref-walk may continue from there. If this is not successful
635 * (eg. a seqcount has changed), then failure is returned and it's up to caller
636 * to restart the path walk from the beginning in ref-walk mode.
640 * unlazy_walk - try to switch to ref-walk mode.
641 * @nd: nameidata pathwalk data
642 * @dentry: child of nd->path.dentry or NULL
643 * @seq: seq number to check dentry against
644 * Returns: 0 on success, -ECHILD on failure
646 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
647 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
648 * @nd or NULL. Must be called from rcu-walk context.
649 * Nothing should touch nameidata between unlazy_walk() failure and
652 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
654 struct dentry *parent = nd->path.dentry;
656 BUG_ON(!(nd->flags & LOOKUP_RCU));
658 nd->flags &= ~LOOKUP_RCU;
659 if (unlikely(!legitimize_links(nd)))
661 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
663 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
667 * For a negative lookup, the lookup sequence point is the parents
668 * sequence point, and it only needs to revalidate the parent dentry.
670 * For a positive lookup, we need to move both the parent and the
671 * dentry from the RCU domain to be properly refcounted. And the
672 * sequence number in the dentry validates *both* dentry counters,
673 * since we checked the sequence number of the parent after we got
674 * the child sequence number. So we know the parent must still
675 * be valid if the child sequence number is still valid.
678 if (read_seqcount_retry(&parent->d_seq, nd->seq))
680 BUG_ON(nd->inode != parent->d_inode);
682 if (!lockref_get_not_dead(&dentry->d_lockref))
684 if (read_seqcount_retry(&dentry->d_seq, seq))
689 * Sequence counts matched. Now make sure that the root is
690 * still valid and get it if required.
692 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
693 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
710 nd->path.dentry = NULL;
714 if (!(nd->flags & LOOKUP_ROOT))
719 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
721 if (unlikely(!legitimize_path(nd, link, seq))) {
724 nd->flags &= ~LOOKUP_RCU;
726 nd->path.dentry = NULL;
727 if (!(nd->flags & LOOKUP_ROOT))
730 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
737 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
739 return dentry->d_op->d_revalidate(dentry, flags);
743 * complete_walk - successful completion of path walk
744 * @nd: pointer nameidata
746 * If we had been in RCU mode, drop out of it and legitimize nd->path.
747 * Revalidate the final result, unless we'd already done that during
748 * the path walk or the filesystem doesn't ask for it. Return 0 on
749 * success, -error on failure. In case of failure caller does not
750 * need to drop nd->path.
752 static int complete_walk(struct nameidata *nd)
754 struct dentry *dentry = nd->path.dentry;
757 if (nd->flags & LOOKUP_RCU) {
758 if (!(nd->flags & LOOKUP_ROOT))
760 if (unlikely(unlazy_walk(nd, NULL, 0)))
764 if (likely(!(nd->flags & LOOKUP_JUMPED)))
767 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
770 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
780 static __always_inline void set_root(struct nameidata *nd)
782 get_fs_root(current->fs, &nd->root);
785 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
787 struct fs_struct *fs = current->fs;
791 seq = read_seqcount_begin(&fs->seq);
793 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
794 } while (read_seqcount_retry(&fs->seq, seq));
798 static void path_put_conditional(struct path *path, struct nameidata *nd)
801 if (path->mnt != nd->path.mnt)
805 static inline void path_to_nameidata(const struct path *path,
806 struct nameidata *nd)
808 if (!(nd->flags & LOOKUP_RCU)) {
809 dput(nd->path.dentry);
810 if (nd->path.mnt != path->mnt)
811 mntput(nd->path.mnt);
813 nd->path.mnt = path->mnt;
814 nd->path.dentry = path->dentry;
818 * Helper to directly jump to a known parsed path from ->follow_link,
819 * caller must have taken a reference to path beforehand.
821 void nd_jump_link(struct path *path)
823 struct nameidata *nd = current->nameidata;
827 nd->inode = nd->path.dentry->d_inode;
828 nd->flags |= LOOKUP_JUMPED;
831 static inline void put_link(struct nameidata *nd)
833 struct saved *last = nd->stack + --nd->depth;
834 struct inode *inode = last->inode;
835 if (last->cookie && inode->i_op->put_link)
836 inode->i_op->put_link(inode, last->cookie);
837 if (!(nd->flags & LOOKUP_RCU))
838 path_put(&last->link);
841 int sysctl_protected_symlinks __read_mostly = 0;
842 int sysctl_protected_hardlinks __read_mostly = 0;
845 * may_follow_link - Check symlink following for unsafe situations
846 * @nd: nameidata pathwalk data
848 * In the case of the sysctl_protected_symlinks sysctl being enabled,
849 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
850 * in a sticky world-writable directory. This is to protect privileged
851 * processes from failing races against path names that may change out
852 * from under them by way of other users creating malicious symlinks.
853 * It will permit symlinks to be followed only when outside a sticky
854 * world-writable directory, or when the uid of the symlink and follower
855 * match, or when the directory owner matches the symlink's owner.
857 * Returns 0 if following the symlink is allowed, -ve on error.
859 static inline int may_follow_link(struct nameidata *nd)
861 const struct inode *inode;
862 const struct inode *parent;
864 if (!sysctl_protected_symlinks)
867 /* Allowed if owner and follower match. */
868 inode = nd->stack[0].inode;
869 if (uid_eq(current_cred()->fsuid, inode->i_uid))
872 /* Allowed if parent directory not sticky and world-writable. */
873 parent = nd->path.dentry->d_inode;
874 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
877 /* Allowed if parent directory and link owner match. */
878 if (uid_eq(parent->i_uid, inode->i_uid))
881 if (nd->flags & LOOKUP_RCU)
884 audit_log_link_denied("follow_link", &nd->stack[0].link);
889 * safe_hardlink_source - Check for safe hardlink conditions
890 * @inode: the source inode to hardlink from
892 * Return false if at least one of the following conditions:
893 * - inode is not a regular file
895 * - inode is setgid and group-exec
896 * - access failure for read and write
898 * Otherwise returns true.
900 static bool safe_hardlink_source(struct inode *inode)
902 umode_t mode = inode->i_mode;
904 /* Special files should not get pinned to the filesystem. */
908 /* Setuid files should not get pinned to the filesystem. */
912 /* Executable setgid files should not get pinned to the filesystem. */
913 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
916 /* Hardlinking to unreadable or unwritable sources is dangerous. */
917 if (inode_permission(inode, MAY_READ | MAY_WRITE))
924 * may_linkat - Check permissions for creating a hardlink
925 * @link: the source to hardlink from
927 * Block hardlink when all of:
928 * - sysctl_protected_hardlinks enabled
929 * - fsuid does not match inode
930 * - hardlink source is unsafe (see safe_hardlink_source() above)
933 * Returns 0 if successful, -ve on error.
935 static int may_linkat(struct path *link)
937 const struct cred *cred;
940 if (!sysctl_protected_hardlinks)
943 cred = current_cred();
944 inode = link->dentry->d_inode;
946 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
947 * otherwise, it must be a safe source.
949 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
953 audit_log_link_denied("linkat", link);
957 static __always_inline
958 const char *get_link(struct nameidata *nd)
960 struct saved *last = nd->stack + nd->depth - 1;
961 struct dentry *dentry = last->link.dentry;
962 struct inode *inode = last->inode;
966 if (!(nd->flags & LOOKUP_RCU)) {
967 touch_atime(&last->link);
969 } else if (atime_needs_update(&last->link, inode)) {
970 if (unlikely(unlazy_walk(nd, NULL, 0)))
971 return ERR_PTR(-ECHILD);
972 touch_atime(&last->link);
975 error = security_inode_follow_link(dentry, inode,
976 nd->flags & LOOKUP_RCU);
978 return ERR_PTR(error);
980 nd->last_type = LAST_BIND;
983 if (nd->flags & LOOKUP_RCU) {
984 if (unlikely(unlazy_walk(nd, NULL, 0)))
985 return ERR_PTR(-ECHILD);
987 res = inode->i_op->follow_link(dentry, &last->cookie);
988 if (IS_ERR_OR_NULL(res)) {
994 if (nd->flags & LOOKUP_RCU) {
1000 nd->inode = d->d_inode;
1001 nd->seq = nd->root_seq;
1002 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1003 return ERR_PTR(-ECHILD);
1007 path_put(&nd->path);
1008 nd->path = nd->root;
1009 path_get(&nd->root);
1010 nd->inode = nd->path.dentry->d_inode;
1012 nd->flags |= LOOKUP_JUMPED;
1013 while (unlikely(*++res == '/'))
1021 static int follow_up_rcu(struct path *path)
1023 struct mount *mnt = real_mount(path->mnt);
1024 struct mount *parent;
1025 struct dentry *mountpoint;
1027 parent = mnt->mnt_parent;
1028 if (&parent->mnt == path->mnt)
1030 mountpoint = mnt->mnt_mountpoint;
1031 path->dentry = mountpoint;
1032 path->mnt = &parent->mnt;
1037 * follow_up - Find the mountpoint of path's vfsmount
1039 * Given a path, find the mountpoint of its source file system.
1040 * Replace @path with the path of the mountpoint in the parent mount.
1043 * Return 1 if we went up a level and 0 if we were already at the
1046 int follow_up(struct path *path)
1048 struct mount *mnt = real_mount(path->mnt);
1049 struct mount *parent;
1050 struct dentry *mountpoint;
1052 read_seqlock_excl(&mount_lock);
1053 parent = mnt->mnt_parent;
1054 if (parent == mnt) {
1055 read_sequnlock_excl(&mount_lock);
1058 mntget(&parent->mnt);
1059 mountpoint = dget(mnt->mnt_mountpoint);
1060 read_sequnlock_excl(&mount_lock);
1062 path->dentry = mountpoint;
1064 path->mnt = &parent->mnt;
1067 EXPORT_SYMBOL(follow_up);
1070 * Perform an automount
1071 * - return -EISDIR to tell follow_managed() to stop and return the path we
1074 static int follow_automount(struct path *path, struct nameidata *nd,
1077 struct vfsmount *mnt;
1080 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1083 /* We don't want to mount if someone's just doing a stat -
1084 * unless they're stat'ing a directory and appended a '/' to
1087 * We do, however, want to mount if someone wants to open or
1088 * create a file of any type under the mountpoint, wants to
1089 * traverse through the mountpoint or wants to open the
1090 * mounted directory. Also, autofs may mark negative dentries
1091 * as being automount points. These will need the attentions
1092 * of the daemon to instantiate them before they can be used.
1094 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1095 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1096 path->dentry->d_inode)
1099 nd->total_link_count++;
1100 if (nd->total_link_count >= 40)
1103 mnt = path->dentry->d_op->d_automount(path);
1106 * The filesystem is allowed to return -EISDIR here to indicate
1107 * it doesn't want to automount. For instance, autofs would do
1108 * this so that its userspace daemon can mount on this dentry.
1110 * However, we can only permit this if it's a terminal point in
1111 * the path being looked up; if it wasn't then the remainder of
1112 * the path is inaccessible and we should say so.
1114 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1116 return PTR_ERR(mnt);
1119 if (!mnt) /* mount collision */
1122 if (!*need_mntput) {
1123 /* lock_mount() may release path->mnt on error */
1125 *need_mntput = true;
1127 err = finish_automount(mnt, path);
1131 /* Someone else made a mount here whilst we were busy */
1136 path->dentry = dget(mnt->mnt_root);
1145 * Handle a dentry that is managed in some way.
1146 * - Flagged for transit management (autofs)
1147 * - Flagged as mountpoint
1148 * - Flagged as automount point
1150 * This may only be called in refwalk mode.
1152 * Serialization is taken care of in namespace.c
1154 static int follow_managed(struct path *path, struct nameidata *nd)
1156 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1158 bool need_mntput = false;
1161 /* Given that we're not holding a lock here, we retain the value in a
1162 * local variable for each dentry as we look at it so that we don't see
1163 * the components of that value change under us */
1164 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1165 managed &= DCACHE_MANAGED_DENTRY,
1166 unlikely(managed != 0)) {
1167 /* Allow the filesystem to manage the transit without i_mutex
1169 if (managed & DCACHE_MANAGE_TRANSIT) {
1170 BUG_ON(!path->dentry->d_op);
1171 BUG_ON(!path->dentry->d_op->d_manage);
1172 ret = path->dentry->d_op->d_manage(path->dentry, false);
1177 /* Transit to a mounted filesystem. */
1178 if (managed & DCACHE_MOUNTED) {
1179 struct vfsmount *mounted = lookup_mnt(path);
1184 path->mnt = mounted;
1185 path->dentry = dget(mounted->mnt_root);
1190 /* Something is mounted on this dentry in another
1191 * namespace and/or whatever was mounted there in this
1192 * namespace got unmounted before lookup_mnt() could
1196 /* Handle an automount point */
1197 if (managed & DCACHE_NEED_AUTOMOUNT) {
1198 ret = follow_automount(path, nd, &need_mntput);
1204 /* We didn't change the current path point */
1208 if (need_mntput && path->mnt == mnt)
1213 nd->flags |= LOOKUP_JUMPED;
1214 if (unlikely(ret < 0))
1215 path_put_conditional(path, nd);
1219 int follow_down_one(struct path *path)
1221 struct vfsmount *mounted;
1223 mounted = lookup_mnt(path);
1227 path->mnt = mounted;
1228 path->dentry = dget(mounted->mnt_root);
1233 EXPORT_SYMBOL(follow_down_one);
1235 static inline int managed_dentry_rcu(struct dentry *dentry)
1237 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1238 dentry->d_op->d_manage(dentry, true) : 0;
1242 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1243 * we meet a managed dentry that would need blocking.
1245 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1246 struct inode **inode, unsigned *seqp)
1249 struct mount *mounted;
1251 * Don't forget we might have a non-mountpoint managed dentry
1252 * that wants to block transit.
1254 switch (managed_dentry_rcu(path->dentry)) {
1264 if (!d_mountpoint(path->dentry))
1265 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1267 mounted = __lookup_mnt(path->mnt, path->dentry);
1270 path->mnt = &mounted->mnt;
1271 path->dentry = mounted->mnt.mnt_root;
1272 nd->flags |= LOOKUP_JUMPED;
1273 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1275 * Update the inode too. We don't need to re-check the
1276 * dentry sequence number here after this d_inode read,
1277 * because a mount-point is always pinned.
1279 *inode = path->dentry->d_inode;
1281 return !read_seqretry(&mount_lock, nd->m_seq) &&
1282 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1285 static int follow_dotdot_rcu(struct nameidata *nd)
1287 struct inode *inode = nd->inode;
1292 if (nd->path.dentry == nd->root.dentry &&
1293 nd->path.mnt == nd->root.mnt) {
1296 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1297 struct dentry *old = nd->path.dentry;
1298 struct dentry *parent = old->d_parent;
1301 inode = parent->d_inode;
1302 seq = read_seqcount_begin(&parent->d_seq);
1303 if (read_seqcount_retry(&old->d_seq, nd->seq))
1305 nd->path.dentry = parent;
1309 if (!follow_up_rcu(&nd->path))
1311 inode = nd->path.dentry->d_inode;
1312 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1314 while (d_mountpoint(nd->path.dentry)) {
1315 struct mount *mounted;
1316 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1319 nd->path.mnt = &mounted->mnt;
1320 nd->path.dentry = mounted->mnt.mnt_root;
1321 inode = nd->path.dentry->d_inode;
1322 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1323 if (read_seqretry(&mount_lock, nd->m_seq))
1334 * Follow down to the covering mount currently visible to userspace. At each
1335 * point, the filesystem owning that dentry may be queried as to whether the
1336 * caller is permitted to proceed or not.
1338 int follow_down(struct path *path)
1343 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1344 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1345 /* Allow the filesystem to manage the transit without i_mutex
1348 * We indicate to the filesystem if someone is trying to mount
1349 * something here. This gives autofs the chance to deny anyone
1350 * other than its daemon the right to mount on its
1353 * The filesystem may sleep at this point.
1355 if (managed & DCACHE_MANAGE_TRANSIT) {
1356 BUG_ON(!path->dentry->d_op);
1357 BUG_ON(!path->dentry->d_op->d_manage);
1358 ret = path->dentry->d_op->d_manage(
1359 path->dentry, false);
1361 return ret == -EISDIR ? 0 : ret;
1364 /* Transit to a mounted filesystem. */
1365 if (managed & DCACHE_MOUNTED) {
1366 struct vfsmount *mounted = lookup_mnt(path);
1371 path->mnt = mounted;
1372 path->dentry = dget(mounted->mnt_root);
1376 /* Don't handle automount points here */
1381 EXPORT_SYMBOL(follow_down);
1384 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1386 static void follow_mount(struct path *path)
1388 while (d_mountpoint(path->dentry)) {
1389 struct vfsmount *mounted = lookup_mnt(path);
1394 path->mnt = mounted;
1395 path->dentry = dget(mounted->mnt_root);
1399 static void follow_dotdot(struct nameidata *nd)
1405 struct dentry *old = nd->path.dentry;
1407 if (nd->path.dentry == nd->root.dentry &&
1408 nd->path.mnt == nd->root.mnt) {
1411 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1412 /* rare case of legitimate dget_parent()... */
1413 nd->path.dentry = dget_parent(nd->path.dentry);
1417 if (!follow_up(&nd->path))
1420 follow_mount(&nd->path);
1421 nd->inode = nd->path.dentry->d_inode;
1425 * This looks up the name in dcache, possibly revalidates the old dentry and
1426 * allocates a new one if not found or not valid. In the need_lookup argument
1427 * returns whether i_op->lookup is necessary.
1429 * dir->d_inode->i_mutex must be held
1431 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1432 unsigned int flags, bool *need_lookup)
1434 struct dentry *dentry;
1437 *need_lookup = false;
1438 dentry = d_lookup(dir, name);
1440 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1441 error = d_revalidate(dentry, flags);
1442 if (unlikely(error <= 0)) {
1445 return ERR_PTR(error);
1447 d_invalidate(dentry);
1456 dentry = d_alloc(dir, name);
1457 if (unlikely(!dentry))
1458 return ERR_PTR(-ENOMEM);
1460 *need_lookup = true;
1466 * Call i_op->lookup on the dentry. The dentry must be negative and
1469 * dir->d_inode->i_mutex must be held
1471 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1476 /* Don't create child dentry for a dead directory. */
1477 if (unlikely(IS_DEADDIR(dir))) {
1479 return ERR_PTR(-ENOENT);
1482 old = dir->i_op->lookup(dir, dentry, flags);
1483 if (unlikely(old)) {
1490 static struct dentry *__lookup_hash(struct qstr *name,
1491 struct dentry *base, unsigned int flags)
1494 struct dentry *dentry;
1496 dentry = lookup_dcache(name, base, flags, &need_lookup);
1500 return lookup_real(base->d_inode, dentry, flags);
1504 * It's more convoluted than I'd like it to be, but... it's still fairly
1505 * small and for now I'd prefer to have fast path as straight as possible.
1506 * It _is_ time-critical.
1508 static int lookup_fast(struct nameidata *nd,
1509 struct path *path, struct inode **inode,
1512 struct vfsmount *mnt = nd->path.mnt;
1513 struct dentry *dentry, *parent = nd->path.dentry;
1519 * Rename seqlock is not required here because in the off chance
1520 * of a false negative due to a concurrent rename, we're going to
1521 * do the non-racy lookup, below.
1523 if (nd->flags & LOOKUP_RCU) {
1526 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1531 * This sequence count validates that the inode matches
1532 * the dentry name information from lookup.
1534 *inode = d_backing_inode(dentry);
1535 negative = d_is_negative(dentry);
1536 if (read_seqcount_retry(&dentry->d_seq, seq))
1542 * This sequence count validates that the parent had no
1543 * changes while we did the lookup of the dentry above.
1545 * The memory barrier in read_seqcount_begin of child is
1546 * enough, we can use __read_seqcount_retry here.
1548 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1552 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1553 status = d_revalidate(dentry, nd->flags);
1554 if (unlikely(status <= 0)) {
1555 if (status != -ECHILD)
1561 path->dentry = dentry;
1562 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1565 if (unlazy_walk(nd, dentry, seq))
1568 dentry = __d_lookup(parent, &nd->last);
1571 if (unlikely(!dentry))
1574 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1575 status = d_revalidate(dentry, nd->flags);
1576 if (unlikely(status <= 0)) {
1581 d_invalidate(dentry);
1586 if (unlikely(d_is_negative(dentry))) {
1591 path->dentry = dentry;
1592 err = follow_managed(path, nd);
1594 *inode = d_backing_inode(path->dentry);
1601 /* Fast lookup failed, do it the slow way */
1602 static int lookup_slow(struct nameidata *nd, struct path *path)
1604 struct dentry *dentry, *parent;
1606 parent = nd->path.dentry;
1607 BUG_ON(nd->inode != parent->d_inode);
1609 mutex_lock(&parent->d_inode->i_mutex);
1610 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1611 mutex_unlock(&parent->d_inode->i_mutex);
1613 return PTR_ERR(dentry);
1614 path->mnt = nd->path.mnt;
1615 path->dentry = dentry;
1616 return follow_managed(path, nd);
1619 static inline int may_lookup(struct nameidata *nd)
1621 if (nd->flags & LOOKUP_RCU) {
1622 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1625 if (unlazy_walk(nd, NULL, 0))
1628 return inode_permission(nd->inode, MAY_EXEC);
1631 static inline int handle_dots(struct nameidata *nd, int type)
1633 if (type == LAST_DOTDOT) {
1634 if (nd->flags & LOOKUP_RCU) {
1635 return follow_dotdot_rcu(nd);
1642 static int pick_link(struct nameidata *nd, struct path *link,
1643 struct inode *inode, unsigned seq)
1647 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1648 path_to_nameidata(link, nd);
1651 if (!(nd->flags & LOOKUP_RCU)) {
1652 if (link->mnt == nd->path.mnt)
1655 error = nd_alloc_stack(nd);
1656 if (unlikely(error)) {
1657 if (error == -ECHILD) {
1658 if (unlikely(unlazy_link(nd, link, seq)))
1660 error = nd_alloc_stack(nd);
1668 last = nd->stack + nd->depth++;
1670 last->cookie = NULL;
1671 last->inode = inode;
1677 * Do we need to follow links? We _really_ want to be able
1678 * to do this check without having to look at inode->i_op,
1679 * so we keep a cache of "no, this doesn't need follow_link"
1680 * for the common case.
1682 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1684 struct inode *inode, unsigned seq)
1686 if (likely(!d_is_symlink(link->dentry)))
1690 return pick_link(nd, link, inode, seq);
1693 enum {WALK_GET = 1, WALK_PUT = 2};
1695 static int walk_component(struct nameidata *nd, int flags)
1698 struct inode *inode;
1702 * "." and ".." are special - ".." especially so because it has
1703 * to be able to know about the current root directory and
1704 * parent relationships.
1706 if (unlikely(nd->last_type != LAST_NORM)) {
1707 err = handle_dots(nd, nd->last_type);
1708 if (flags & WALK_PUT)
1712 err = lookup_fast(nd, &path, &inode, &seq);
1713 if (unlikely(err)) {
1717 err = lookup_slow(nd, &path);
1721 inode = d_backing_inode(path.dentry);
1722 seq = 0; /* we are already out of RCU mode */
1724 if (d_is_negative(path.dentry))
1728 if (flags & WALK_PUT)
1730 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1733 path_to_nameidata(&path, nd);
1739 path_to_nameidata(&path, nd);
1744 * We can do the critical dentry name comparison and hashing
1745 * operations one word at a time, but we are limited to:
1747 * - Architectures with fast unaligned word accesses. We could
1748 * do a "get_unaligned()" if this helps and is sufficiently
1751 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1752 * do not trap on the (extremely unlikely) case of a page
1753 * crossing operation.
1755 * - Furthermore, we need an efficient 64-bit compile for the
1756 * 64-bit case in order to generate the "number of bytes in
1757 * the final mask". Again, that could be replaced with a
1758 * efficient population count instruction or similar.
1760 #ifdef CONFIG_DCACHE_WORD_ACCESS
1762 #include <asm/word-at-a-time.h>
1766 static inline unsigned int fold_hash(unsigned long hash)
1768 return hash_64(hash, 32);
1771 #else /* 32-bit case */
1773 #define fold_hash(x) (x)
1777 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1779 unsigned long a, mask;
1780 unsigned long hash = 0;
1783 a = load_unaligned_zeropad(name);
1784 if (len < sizeof(unsigned long))
1788 name += sizeof(unsigned long);
1789 len -= sizeof(unsigned long);
1793 mask = bytemask_from_count(len);
1796 return fold_hash(hash);
1798 EXPORT_SYMBOL(full_name_hash);
1801 * Calculate the length and hash of the path component, and
1802 * return the "hash_len" as the result.
1804 static inline u64 hash_name(const char *name)
1806 unsigned long a, b, adata, bdata, mask, hash, len;
1807 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1810 len = -sizeof(unsigned long);
1812 hash = (hash + a) * 9;
1813 len += sizeof(unsigned long);
1814 a = load_unaligned_zeropad(name+len);
1815 b = a ^ REPEAT_BYTE('/');
1816 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1818 adata = prep_zero_mask(a, adata, &constants);
1819 bdata = prep_zero_mask(b, bdata, &constants);
1821 mask = create_zero_mask(adata | bdata);
1823 hash += a & zero_bytemask(mask);
1824 len += find_zero(mask);
1825 return hashlen_create(fold_hash(hash), len);
1830 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1832 unsigned long hash = init_name_hash();
1834 hash = partial_name_hash(*name++, hash);
1835 return end_name_hash(hash);
1837 EXPORT_SYMBOL(full_name_hash);
1840 * We know there's a real path component here of at least
1843 static inline u64 hash_name(const char *name)
1845 unsigned long hash = init_name_hash();
1846 unsigned long len = 0, c;
1848 c = (unsigned char)*name;
1851 hash = partial_name_hash(c, hash);
1852 c = (unsigned char)name[len];
1853 } while (c && c != '/');
1854 return hashlen_create(end_name_hash(hash), len);
1861 * This is the basic name resolution function, turning a pathname into
1862 * the final dentry. We expect 'base' to be positive and a directory.
1864 * Returns 0 and nd will have valid dentry and mnt on success.
1865 * Returns error and drops reference to input namei data on failure.
1867 static int link_path_walk(const char *name, struct nameidata *nd)
1876 /* At this point we know we have a real path component. */
1881 err = may_lookup(nd);
1885 hash_len = hash_name(name);
1888 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1890 if (name[1] == '.') {
1892 nd->flags |= LOOKUP_JUMPED;
1898 if (likely(type == LAST_NORM)) {
1899 struct dentry *parent = nd->path.dentry;
1900 nd->flags &= ~LOOKUP_JUMPED;
1901 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1902 struct qstr this = { { .hash_len = hash_len }, .name = name };
1903 err = parent->d_op->d_hash(parent, &this);
1906 hash_len = this.hash_len;
1911 nd->last.hash_len = hash_len;
1912 nd->last.name = name;
1913 nd->last_type = type;
1915 name += hashlen_len(hash_len);
1919 * If it wasn't NUL, we know it was '/'. Skip that
1920 * slash, and continue until no more slashes.
1924 } while (unlikely(*name == '/'));
1925 if (unlikely(!*name)) {
1927 /* pathname body, done */
1930 name = nd->stack[nd->depth - 1].name;
1931 /* trailing symlink, done */
1934 /* last component of nested symlink */
1935 err = walk_component(nd, WALK_GET | WALK_PUT);
1937 err = walk_component(nd, WALK_GET);
1943 const char *s = get_link(nd);
1945 if (unlikely(IS_ERR(s)))
1952 nd->stack[nd->depth - 1].name = name;
1957 if (unlikely(!d_can_lookup(nd->path.dentry)))
1962 static const char *path_init(int dfd, const struct filename *name,
1963 unsigned int flags, struct nameidata *nd)
1966 const char *s = name->name;
1968 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1969 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1971 nd->total_link_count = 0;
1972 if (flags & LOOKUP_ROOT) {
1973 struct dentry *root = nd->root.dentry;
1974 struct inode *inode = root->d_inode;
1976 if (!d_can_lookup(root))
1977 return ERR_PTR(-ENOTDIR);
1978 retval = inode_permission(inode, MAY_EXEC);
1980 return ERR_PTR(retval);
1982 nd->path = nd->root;
1984 if (flags & LOOKUP_RCU) {
1986 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1987 nd->root_seq = nd->seq;
1988 nd->m_seq = read_seqbegin(&mount_lock);
1990 path_get(&nd->path);
1995 nd->root.mnt = NULL;
1997 nd->m_seq = read_seqbegin(&mount_lock);
1999 if (flags & LOOKUP_RCU) {
2001 nd->seq = set_root_rcu(nd);
2004 path_get(&nd->root);
2006 nd->path = nd->root;
2007 } else if (dfd == AT_FDCWD) {
2008 if (flags & LOOKUP_RCU) {
2009 struct fs_struct *fs = current->fs;
2015 seq = read_seqcount_begin(&fs->seq);
2017 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2018 } while (read_seqcount_retry(&fs->seq, seq));
2020 get_fs_pwd(current->fs, &nd->path);
2023 /* Caller must check execute permissions on the starting path component */
2024 struct fd f = fdget_raw(dfd);
2025 struct dentry *dentry;
2028 return ERR_PTR(-EBADF);
2030 dentry = f.file->f_path.dentry;
2033 if (!d_can_lookup(dentry)) {
2035 return ERR_PTR(-ENOTDIR);
2039 nd->path = f.file->f_path;
2040 if (flags & LOOKUP_RCU) {
2042 nd->inode = nd->path.dentry->d_inode;
2043 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2045 path_get(&nd->path);
2046 nd->inode = nd->path.dentry->d_inode;
2052 nd->inode = nd->path.dentry->d_inode;
2053 if (!(flags & LOOKUP_RCU))
2055 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2057 if (!(nd->flags & LOOKUP_ROOT))
2058 nd->root.mnt = NULL;
2060 return ERR_PTR(-ECHILD);
2063 static void path_cleanup(struct nameidata *nd)
2065 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2066 path_put(&nd->root);
2067 nd->root.mnt = NULL;
2071 static const char *trailing_symlink(struct nameidata *nd)
2074 int error = may_follow_link(nd);
2075 if (unlikely(error))
2076 return ERR_PTR(error);
2077 nd->flags |= LOOKUP_PARENT;
2078 nd->stack[0].name = NULL;
2083 static inline int lookup_last(struct nameidata *nd)
2085 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2086 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2088 nd->flags &= ~LOOKUP_PARENT;
2089 return walk_component(nd,
2090 nd->flags & LOOKUP_FOLLOW
2092 ? WALK_PUT | WALK_GET
2097 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2098 static int path_lookupat(int dfd, const struct filename *name,
2099 unsigned int flags, struct nameidata *nd)
2101 const char *s = path_init(dfd, name, flags, nd);
2106 while (!(err = link_path_walk(s, nd))
2107 && ((err = lookup_last(nd)) > 0)) {
2108 s = trailing_symlink(nd);
2115 err = complete_walk(nd);
2117 if (!err && nd->flags & LOOKUP_DIRECTORY)
2118 if (!d_can_lookup(nd->path.dentry))
2127 static int filename_lookup(int dfd, struct filename *name,
2128 unsigned int flags, struct nameidata *nd)
2131 struct nameidata *saved_nd = set_nameidata(nd);
2133 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2134 if (unlikely(retval == -ECHILD))
2135 retval = path_lookupat(dfd, name, flags, nd);
2136 if (unlikely(retval == -ESTALE))
2137 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2139 if (likely(!retval))
2140 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2141 restore_nameidata(saved_nd);
2145 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2146 static int path_parentat(int dfd, const struct filename *name,
2147 unsigned int flags, struct nameidata *nd)
2149 const char *s = path_init(dfd, name, flags, nd);
2153 err = link_path_walk(s, nd);
2155 err = complete_walk(nd);
2162 static int filename_parentat(int dfd, struct filename *name,
2163 unsigned int flags, struct nameidata *nd)
2166 struct nameidata *saved_nd = set_nameidata(nd);
2168 retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd);
2169 if (unlikely(retval == -ECHILD))
2170 retval = path_parentat(dfd, name, flags, nd);
2171 if (unlikely(retval == -ESTALE))
2172 retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd);
2174 if (likely(!retval))
2175 audit_inode(name, nd->path.dentry, LOOKUP_PARENT);
2176 restore_nameidata(saved_nd);
2180 /* does lookup, returns the object with parent locked */
2181 struct dentry *kern_path_locked(const char *name, struct path *path)
2183 struct filename *filename = getname_kernel(name);
2184 struct nameidata nd;
2188 if (IS_ERR(filename))
2189 return ERR_CAST(filename);
2191 err = filename_parentat(AT_FDCWD, filename, 0, &nd);
2196 if (nd.last_type != LAST_NORM) {
2198 d = ERR_PTR(-EINVAL);
2201 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2202 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2204 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2214 int kern_path(const char *name, unsigned int flags, struct path *path)
2216 struct nameidata nd;
2217 struct filename *filename = getname_kernel(name);
2218 int res = PTR_ERR(filename);
2220 if (!IS_ERR(filename)) {
2221 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2228 EXPORT_SYMBOL(kern_path);
2231 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2232 * @dentry: pointer to dentry of the base directory
2233 * @mnt: pointer to vfs mount of the base directory
2234 * @name: pointer to file name
2235 * @flags: lookup flags
2236 * @path: pointer to struct path to fill
2238 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2239 const char *name, unsigned int flags,
2242 struct filename *filename = getname_kernel(name);
2243 int err = PTR_ERR(filename);
2245 BUG_ON(flags & LOOKUP_PARENT);
2247 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2248 if (!IS_ERR(filename)) {
2249 struct nameidata nd;
2250 nd.root.dentry = dentry;
2252 err = filename_lookup(AT_FDCWD, filename,
2253 flags | LOOKUP_ROOT, &nd);
2260 EXPORT_SYMBOL(vfs_path_lookup);
2263 * lookup_one_len - filesystem helper to lookup single pathname component
2264 * @name: pathname component to lookup
2265 * @base: base directory to lookup from
2266 * @len: maximum length @len should be interpreted to
2268 * Note that this routine is purely a helper for filesystem usage and should
2269 * not be called by generic code.
2271 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2277 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2281 this.hash = full_name_hash(name, len);
2283 return ERR_PTR(-EACCES);
2285 if (unlikely(name[0] == '.')) {
2286 if (len < 2 || (len == 2 && name[1] == '.'))
2287 return ERR_PTR(-EACCES);
2291 c = *(const unsigned char *)name++;
2292 if (c == '/' || c == '\0')
2293 return ERR_PTR(-EACCES);
2296 * See if the low-level filesystem might want
2297 * to use its own hash..
2299 if (base->d_flags & DCACHE_OP_HASH) {
2300 int err = base->d_op->d_hash(base, &this);
2302 return ERR_PTR(err);
2305 err = inode_permission(base->d_inode, MAY_EXEC);
2307 return ERR_PTR(err);
2309 return __lookup_hash(&this, base, 0);
2311 EXPORT_SYMBOL(lookup_one_len);
2313 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2314 struct path *path, int *empty)
2316 struct nameidata nd;
2317 struct filename *tmp = getname_flags(name, flags, empty);
2318 int err = PTR_ERR(tmp);
2321 BUG_ON(flags & LOOKUP_PARENT);
2323 err = filename_lookup(dfd, tmp, flags, &nd);
2331 int user_path_at(int dfd, const char __user *name, unsigned flags,
2334 return user_path_at_empty(dfd, name, flags, path, NULL);
2336 EXPORT_SYMBOL(user_path_at);
2339 * NB: most callers don't do anything directly with the reference to the
2340 * to struct filename, but the nd->last pointer points into the name string
2341 * allocated by getname. So we must hold the reference to it until all
2342 * path-walking is complete.
2344 static struct filename *
2345 user_path_parent(int dfd, const char __user *path,
2346 struct path *parent,
2351 struct nameidata nd;
2352 struct filename *s = getname(path);
2355 /* only LOOKUP_REVAL is allowed in extra flags */
2356 flags &= LOOKUP_REVAL;
2361 error = filename_parentat(dfd, s, flags, &nd);
2364 return ERR_PTR(error);
2368 *type = nd.last_type;
2374 * mountpoint_last - look up last component for umount
2375 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2376 * @path: pointer to container for result
2378 * This is a special lookup_last function just for umount. In this case, we
2379 * need to resolve the path without doing any revalidation.
2381 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2382 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2383 * in almost all cases, this lookup will be served out of the dcache. The only
2384 * cases where it won't are if nd->last refers to a symlink or the path is
2385 * bogus and it doesn't exist.
2388 * -error: if there was an error during lookup. This includes -ENOENT if the
2389 * lookup found a negative dentry. The nd->path reference will also be
2392 * 0: if we successfully resolved nd->path and found it to not to be a
2393 * symlink that needs to be followed. "path" will also be populated.
2394 * The nd->path reference will also be put.
2396 * 1: if we successfully resolved nd->last and found it to be a symlink
2397 * that needs to be followed. "path" will be populated with the path
2398 * to the link, and nd->path will *not* be put.
2401 mountpoint_last(struct nameidata *nd, struct path *path)
2404 struct dentry *dentry;
2405 struct dentry *dir = nd->path.dentry;
2407 /* If we're in rcuwalk, drop out of it to handle last component */
2408 if (nd->flags & LOOKUP_RCU) {
2409 if (unlazy_walk(nd, NULL, 0))
2413 nd->flags &= ~LOOKUP_PARENT;
2415 if (unlikely(nd->last_type != LAST_NORM)) {
2416 error = handle_dots(nd, nd->last_type);
2419 dentry = dget(nd->path.dentry);
2423 mutex_lock(&dir->d_inode->i_mutex);
2424 dentry = d_lookup(dir, &nd->last);
2427 * No cached dentry. Mounted dentries are pinned in the cache,
2428 * so that means that this dentry is probably a symlink or the
2429 * path doesn't actually point to a mounted dentry.
2431 dentry = d_alloc(dir, &nd->last);
2433 mutex_unlock(&dir->d_inode->i_mutex);
2436 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2437 if (IS_ERR(dentry)) {
2438 mutex_unlock(&dir->d_inode->i_mutex);
2439 return PTR_ERR(dentry);
2442 mutex_unlock(&dir->d_inode->i_mutex);
2445 if (d_is_negative(dentry)) {
2451 path->dentry = dentry;
2452 path->mnt = nd->path.mnt;
2453 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2454 d_backing_inode(dentry), 0);
2455 if (unlikely(error))
2463 * path_mountpoint - look up a path to be umounted
2464 * @dfd: directory file descriptor to start walk from
2465 * @name: full pathname to walk
2466 * @path: pointer to container for result
2467 * @flags: lookup flags
2469 * Look up the given name, but don't attempt to revalidate the last component.
2470 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2473 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2474 struct nameidata *nd, unsigned int flags)
2476 const char *s = path_init(dfd, name, flags, nd);
2480 while (!(err = link_path_walk(s, nd)) &&
2481 (err = mountpoint_last(nd, path)) > 0) {
2482 s = trailing_symlink(nd);
2494 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2497 struct nameidata nd, *saved;
2500 return PTR_ERR(name);
2501 saved = set_nameidata(&nd);
2502 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2503 if (unlikely(error == -ECHILD))
2504 error = path_mountpoint(dfd, name, path, &nd, flags);
2505 if (unlikely(error == -ESTALE))
2506 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2508 audit_inode(name, path->dentry, 0);
2509 restore_nameidata(saved);
2515 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2516 * @dfd: directory file descriptor
2517 * @name: pathname from userland
2518 * @flags: lookup flags
2519 * @path: pointer to container to hold result
2521 * A umount is a special case for path walking. We're not actually interested
2522 * in the inode in this situation, and ESTALE errors can be a problem. We
2523 * simply want track down the dentry and vfsmount attached at the mountpoint
2524 * and avoid revalidating the last component.
2526 * Returns 0 and populates "path" on success.
2529 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2532 return filename_mountpoint(dfd, getname(name), path, flags);
2536 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2539 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2541 EXPORT_SYMBOL(kern_path_mountpoint);
2543 int __check_sticky(struct inode *dir, struct inode *inode)
2545 kuid_t fsuid = current_fsuid();
2547 if (uid_eq(inode->i_uid, fsuid))
2549 if (uid_eq(dir->i_uid, fsuid))
2551 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2553 EXPORT_SYMBOL(__check_sticky);
2556 * Check whether we can remove a link victim from directory dir, check
2557 * whether the type of victim is right.
2558 * 1. We can't do it if dir is read-only (done in permission())
2559 * 2. We should have write and exec permissions on dir
2560 * 3. We can't remove anything from append-only dir
2561 * 4. We can't do anything with immutable dir (done in permission())
2562 * 5. If the sticky bit on dir is set we should either
2563 * a. be owner of dir, or
2564 * b. be owner of victim, or
2565 * c. have CAP_FOWNER capability
2566 * 6. If the victim is append-only or immutable we can't do antyhing with
2567 * links pointing to it.
2568 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2569 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2570 * 9. We can't remove a root or mountpoint.
2571 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2572 * nfs_async_unlink().
2574 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2576 struct inode *inode = d_backing_inode(victim);
2579 if (d_is_negative(victim))
2583 BUG_ON(victim->d_parent->d_inode != dir);
2584 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2586 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2592 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2593 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2596 if (!d_is_dir(victim))
2598 if (IS_ROOT(victim))
2600 } else if (d_is_dir(victim))
2602 if (IS_DEADDIR(dir))
2604 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2609 /* Check whether we can create an object with dentry child in directory
2611 * 1. We can't do it if child already exists (open has special treatment for
2612 * this case, but since we are inlined it's OK)
2613 * 2. We can't do it if dir is read-only (done in permission())
2614 * 3. We should have write and exec permissions on dir
2615 * 4. We can't do it if dir is immutable (done in permission())
2617 static inline int may_create(struct inode *dir, struct dentry *child)
2619 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2622 if (IS_DEADDIR(dir))
2624 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2628 * p1 and p2 should be directories on the same fs.
2630 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2635 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2639 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2641 p = d_ancestor(p2, p1);
2643 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2644 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2648 p = d_ancestor(p1, p2);
2650 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2651 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2655 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2656 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2659 EXPORT_SYMBOL(lock_rename);
2661 void unlock_rename(struct dentry *p1, struct dentry *p2)
2663 mutex_unlock(&p1->d_inode->i_mutex);
2665 mutex_unlock(&p2->d_inode->i_mutex);
2666 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2669 EXPORT_SYMBOL(unlock_rename);
2671 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2674 int error = may_create(dir, dentry);
2678 if (!dir->i_op->create)
2679 return -EACCES; /* shouldn't it be ENOSYS? */
2682 error = security_inode_create(dir, dentry, mode);
2685 error = dir->i_op->create(dir, dentry, mode, want_excl);
2687 fsnotify_create(dir, dentry);
2690 EXPORT_SYMBOL(vfs_create);
2692 static int may_open(struct path *path, int acc_mode, int flag)
2694 struct dentry *dentry = path->dentry;
2695 struct inode *inode = dentry->d_inode;
2705 switch (inode->i_mode & S_IFMT) {
2709 if (acc_mode & MAY_WRITE)
2714 if (path->mnt->mnt_flags & MNT_NODEV)
2723 error = inode_permission(inode, acc_mode);
2728 * An append-only file must be opened in append mode for writing.
2730 if (IS_APPEND(inode)) {
2731 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2737 /* O_NOATIME can only be set by the owner or superuser */
2738 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2744 static int handle_truncate(struct file *filp)
2746 struct path *path = &filp->f_path;
2747 struct inode *inode = path->dentry->d_inode;
2748 int error = get_write_access(inode);
2752 * Refuse to truncate files with mandatory locks held on them.
2754 error = locks_verify_locked(filp);
2756 error = security_path_truncate(path);
2758 error = do_truncate(path->dentry, 0,
2759 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2762 put_write_access(inode);
2766 static inline int open_to_namei_flags(int flag)
2768 if ((flag & O_ACCMODE) == 3)
2773 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2775 int error = security_path_mknod(dir, dentry, mode, 0);
2779 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2783 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2787 * Attempt to atomically look up, create and open a file from a negative
2790 * Returns 0 if successful. The file will have been created and attached to
2791 * @file by the filesystem calling finish_open().
2793 * Returns 1 if the file was looked up only or didn't need creating. The
2794 * caller will need to perform the open themselves. @path will have been
2795 * updated to point to the new dentry. This may be negative.
2797 * Returns an error code otherwise.
2799 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2800 struct path *path, struct file *file,
2801 const struct open_flags *op,
2802 bool got_write, bool need_lookup,
2805 struct inode *dir = nd->path.dentry->d_inode;
2806 unsigned open_flag = open_to_namei_flags(op->open_flag);
2810 int create_error = 0;
2811 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2814 BUG_ON(dentry->d_inode);
2816 /* Don't create child dentry for a dead directory. */
2817 if (unlikely(IS_DEADDIR(dir))) {
2823 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2824 mode &= ~current_umask();
2826 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2828 open_flag &= ~O_TRUNC;
2831 * Checking write permission is tricky, bacuse we don't know if we are
2832 * going to actually need it: O_CREAT opens should work as long as the
2833 * file exists. But checking existence breaks atomicity. The trick is
2834 * to check access and if not granted clear O_CREAT from the flags.
2836 * Another problem is returing the "right" error value (e.g. for an
2837 * O_EXCL open we want to return EEXIST not EROFS).
2839 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2840 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2841 if (!(open_flag & O_CREAT)) {
2843 * No O_CREATE -> atomicity not a requirement -> fall
2844 * back to lookup + open
2847 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2848 /* Fall back and fail with the right error */
2849 create_error = -EROFS;
2852 /* No side effects, safe to clear O_CREAT */
2853 create_error = -EROFS;
2854 open_flag &= ~O_CREAT;
2858 if (open_flag & O_CREAT) {
2859 error = may_o_create(&nd->path, dentry, mode);
2861 create_error = error;
2862 if (open_flag & O_EXCL)
2864 open_flag &= ~O_CREAT;
2868 if (nd->flags & LOOKUP_DIRECTORY)
2869 open_flag |= O_DIRECTORY;
2871 file->f_path.dentry = DENTRY_NOT_SET;
2872 file->f_path.mnt = nd->path.mnt;
2873 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2876 if (create_error && error == -ENOENT)
2877 error = create_error;
2881 if (error) { /* returned 1, that is */
2882 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2886 if (file->f_path.dentry) {
2888 dentry = file->f_path.dentry;
2890 if (*opened & FILE_CREATED)
2891 fsnotify_create(dir, dentry);
2892 if (!dentry->d_inode) {
2893 WARN_ON(*opened & FILE_CREATED);
2895 error = create_error;
2899 if (excl && !(*opened & FILE_CREATED)) {
2908 * We didn't have the inode before the open, so check open permission
2911 acc_mode = op->acc_mode;
2912 if (*opened & FILE_CREATED) {
2913 WARN_ON(!(open_flag & O_CREAT));
2914 fsnotify_create(dir, dentry);
2915 acc_mode = MAY_OPEN;
2917 error = may_open(&file->f_path, acc_mode, open_flag);
2927 dentry = lookup_real(dir, dentry, nd->flags);
2929 return PTR_ERR(dentry);
2932 int open_flag = op->open_flag;
2934 error = create_error;
2935 if ((open_flag & O_EXCL)) {
2936 if (!dentry->d_inode)
2938 } else if (!dentry->d_inode) {
2940 } else if ((open_flag & O_TRUNC) &&
2944 /* will fail later, go on to get the right error */
2948 path->dentry = dentry;
2949 path->mnt = nd->path.mnt;
2954 * Look up and maybe create and open the last component.
2956 * Must be called with i_mutex held on parent.
2958 * Returns 0 if the file was successfully atomically created (if necessary) and
2959 * opened. In this case the file will be returned attached to @file.
2961 * Returns 1 if the file was not completely opened at this time, though lookups
2962 * and creations will have been performed and the dentry returned in @path will
2963 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2964 * specified then a negative dentry may be returned.
2966 * An error code is returned otherwise.
2968 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2969 * cleared otherwise prior to returning.
2971 static int lookup_open(struct nameidata *nd, struct path *path,
2973 const struct open_flags *op,
2974 bool got_write, int *opened)
2976 struct dentry *dir = nd->path.dentry;
2977 struct inode *dir_inode = dir->d_inode;
2978 struct dentry *dentry;
2982 *opened &= ~FILE_CREATED;
2983 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2985 return PTR_ERR(dentry);
2987 /* Cached positive dentry: will open in f_op->open */
2988 if (!need_lookup && dentry->d_inode)
2991 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2992 return atomic_open(nd, dentry, path, file, op, got_write,
2993 need_lookup, opened);
2997 BUG_ON(dentry->d_inode);
2999 dentry = lookup_real(dir_inode, dentry, nd->flags);
3001 return PTR_ERR(dentry);
3004 /* Negative dentry, just create the file */
3005 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3006 umode_t mode = op->mode;
3007 if (!IS_POSIXACL(dir->d_inode))
3008 mode &= ~current_umask();
3010 * This write is needed to ensure that a
3011 * rw->ro transition does not occur between
3012 * the time when the file is created and when
3013 * a permanent write count is taken through
3014 * the 'struct file' in finish_open().
3020 *opened |= FILE_CREATED;
3021 error = security_path_mknod(&nd->path, dentry, mode, 0);
3024 error = vfs_create(dir->d_inode, dentry, mode,
3025 nd->flags & LOOKUP_EXCL);
3030 path->dentry = dentry;
3031 path->mnt = nd->path.mnt;
3040 * Handle the last step of open()
3042 static int do_last(struct nameidata *nd,
3043 struct file *file, const struct open_flags *op,
3044 int *opened, struct filename *name)
3046 struct dentry *dir = nd->path.dentry;
3047 int open_flag = op->open_flag;
3048 bool will_truncate = (open_flag & O_TRUNC) != 0;
3049 bool got_write = false;
3050 int acc_mode = op->acc_mode;
3052 struct inode *inode;
3053 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3055 bool retried = false;
3058 nd->flags &= ~LOOKUP_PARENT;
3059 nd->flags |= op->intent;
3061 if (nd->last_type != LAST_NORM) {
3062 error = handle_dots(nd, nd->last_type);
3063 if (unlikely(error))
3068 if (!(open_flag & O_CREAT)) {
3069 if (nd->last.name[nd->last.len])
3070 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3071 /* we _can_ be in RCU mode here */
3072 error = lookup_fast(nd, &path, &inode, &seq);
3079 BUG_ON(nd->inode != dir->d_inode);
3081 /* create side of things */
3083 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3084 * has been cleared when we got to the last component we are
3087 error = complete_walk(nd);
3091 audit_inode(name, dir, LOOKUP_PARENT);
3092 /* trailing slashes? */
3093 if (unlikely(nd->last.name[nd->last.len]))
3098 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3099 error = mnt_want_write(nd->path.mnt);
3103 * do _not_ fail yet - we might not need that or fail with
3104 * a different error; let lookup_open() decide; we'll be
3105 * dropping this one anyway.
3108 mutex_lock(&dir->d_inode->i_mutex);
3109 error = lookup_open(nd, &path, file, op, got_write, opened);
3110 mutex_unlock(&dir->d_inode->i_mutex);
3116 if ((*opened & FILE_CREATED) ||
3117 !S_ISREG(file_inode(file)->i_mode))
3118 will_truncate = false;
3120 audit_inode(name, file->f_path.dentry, 0);
3124 if (*opened & FILE_CREATED) {
3125 /* Don't check for write permission, don't truncate */
3126 open_flag &= ~O_TRUNC;
3127 will_truncate = false;
3128 acc_mode = MAY_OPEN;
3129 path_to_nameidata(&path, nd);
3130 goto finish_open_created;
3134 * create/update audit record if it already exists.
3136 if (d_is_positive(path.dentry))
3137 audit_inode(name, path.dentry, 0);
3140 * If atomic_open() acquired write access it is dropped now due to
3141 * possible mount and symlink following (this might be optimized away if
3145 mnt_drop_write(nd->path.mnt);
3149 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3150 path_to_nameidata(&path, nd);
3154 error = follow_managed(&path, nd);
3155 if (unlikely(error < 0))
3158 BUG_ON(nd->flags & LOOKUP_RCU);
3159 inode = d_backing_inode(path.dentry);
3160 seq = 0; /* out of RCU mode, so the value doesn't matter */
3161 if (unlikely(d_is_negative(path.dentry))) {
3162 path_to_nameidata(&path, nd);
3168 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3170 if (unlikely(error))
3173 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3174 path_to_nameidata(&path, nd);
3178 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3179 path_to_nameidata(&path, nd);
3181 save_parent.dentry = nd->path.dentry;
3182 save_parent.mnt = mntget(path.mnt);
3183 nd->path.dentry = path.dentry;
3188 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3190 error = complete_walk(nd);
3192 path_put(&save_parent);
3195 audit_inode(name, nd->path.dentry, 0);
3197 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3200 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3202 if (!d_is_reg(nd->path.dentry))
3203 will_truncate = false;
3205 if (will_truncate) {
3206 error = mnt_want_write(nd->path.mnt);
3211 finish_open_created:
3212 error = may_open(&nd->path, acc_mode, open_flag);
3216 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3217 error = vfs_open(&nd->path, file, current_cred());
3219 *opened |= FILE_OPENED;
3221 if (error == -EOPENSTALE)
3226 error = open_check_o_direct(file);
3229 error = ima_file_check(file, op->acc_mode, *opened);
3233 if (will_truncate) {
3234 error = handle_truncate(file);
3240 mnt_drop_write(nd->path.mnt);
3241 path_put(&save_parent);
3249 /* If no saved parent or already retried then can't retry */
3250 if (!save_parent.dentry || retried)
3253 BUG_ON(save_parent.dentry != dir);
3254 path_put(&nd->path);
3255 nd->path = save_parent;
3256 nd->inode = dir->d_inode;
3257 save_parent.mnt = NULL;
3258 save_parent.dentry = NULL;
3260 mnt_drop_write(nd->path.mnt);
3267 static int do_tmpfile(int dfd, struct filename *pathname,
3268 struct nameidata *nd, int flags,
3269 const struct open_flags *op,
3270 struct file *file, int *opened)
3272 static const struct qstr name = QSTR_INIT("/", 1);
3273 struct dentry *dentry, *child;
3275 int error = path_lookupat(dfd, pathname,
3276 flags | LOOKUP_DIRECTORY, nd);
3277 if (unlikely(error))
3279 error = mnt_want_write(nd->path.mnt);
3280 if (unlikely(error))
3282 /* we want directory to be writable */
3283 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3286 dentry = nd->path.dentry;
3287 dir = dentry->d_inode;
3288 if (!dir->i_op->tmpfile) {
3289 error = -EOPNOTSUPP;
3292 child = d_alloc(dentry, &name);
3293 if (unlikely(!child)) {
3297 nd->flags &= ~LOOKUP_DIRECTORY;
3298 nd->flags |= op->intent;
3299 dput(nd->path.dentry);
3300 nd->path.dentry = child;
3301 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3304 audit_inode(pathname, nd->path.dentry, 0);
3305 /* Don't check for other permissions, the inode was just created */
3306 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3309 file->f_path.mnt = nd->path.mnt;
3310 error = finish_open(file, nd->path.dentry, NULL, opened);
3313 error = open_check_o_direct(file);
3316 } else if (!(op->open_flag & O_EXCL)) {
3317 struct inode *inode = file_inode(file);
3318 spin_lock(&inode->i_lock);
3319 inode->i_state |= I_LINKABLE;
3320 spin_unlock(&inode->i_lock);
3323 mnt_drop_write(nd->path.mnt);
3325 path_put(&nd->path);
3329 static struct file *path_openat(int dfd, struct filename *pathname,
3330 struct nameidata *nd, const struct open_flags *op, int flags)
3337 file = get_empty_filp();
3341 file->f_flags = op->open_flag;
3343 if (unlikely(file->f_flags & __O_TMPFILE)) {
3344 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3348 s = path_init(dfd, pathname, flags, nd);
3353 while (!(error = link_path_walk(s, nd)) &&
3354 (error = do_last(nd, file, op, &opened, pathname)) > 0) {
3355 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3356 s = trailing_symlink(nd);
3365 if (!(opened & FILE_OPENED)) {
3369 if (unlikely(error)) {
3370 if (error == -EOPENSTALE) {
3371 if (flags & LOOKUP_RCU)
3376 file = ERR_PTR(error);
3381 struct file *do_filp_open(int dfd, struct filename *pathname,
3382 const struct open_flags *op)
3384 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3385 int flags = op->lookup_flags;
3388 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3389 if (unlikely(filp == ERR_PTR(-ECHILD)))
3390 filp = path_openat(dfd, pathname, &nd, op, flags);
3391 if (unlikely(filp == ERR_PTR(-ESTALE)))
3392 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3393 restore_nameidata(saved_nd);
3397 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3398 const char *name, const struct open_flags *op)
3400 struct nameidata nd, *saved_nd;
3402 struct filename *filename;
3403 int flags = op->lookup_flags | LOOKUP_ROOT;
3406 nd.root.dentry = dentry;
3408 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3409 return ERR_PTR(-ELOOP);
3411 filename = getname_kernel(name);
3412 if (unlikely(IS_ERR(filename)))
3413 return ERR_CAST(filename);
3415 saved_nd = set_nameidata(&nd);
3416 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3417 if (unlikely(file == ERR_PTR(-ECHILD)))
3418 file = path_openat(-1, filename, &nd, op, flags);
3419 if (unlikely(file == ERR_PTR(-ESTALE)))
3420 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3421 restore_nameidata(saved_nd);
3426 static struct dentry *filename_create(int dfd, struct filename *name,
3427 struct path *path, unsigned int lookup_flags)
3429 struct dentry *dentry = ERR_PTR(-EEXIST);
3430 struct nameidata nd;
3433 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3436 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3437 * other flags passed in are ignored!
3439 lookup_flags &= LOOKUP_REVAL;
3441 error = filename_parentat(dfd, name, lookup_flags, &nd);
3443 return ERR_PTR(error);
3446 * Yucky last component or no last component at all?
3447 * (foo/., foo/.., /////)
3449 if (nd.last_type != LAST_NORM)
3451 nd.flags &= ~LOOKUP_PARENT;
3452 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3454 /* don't fail immediately if it's r/o, at least try to report other errors */
3455 err2 = mnt_want_write(nd.path.mnt);
3457 * Do the final lookup.
3459 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3460 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3465 if (d_is_positive(dentry))
3469 * Special case - lookup gave negative, but... we had foo/bar/
3470 * From the vfs_mknod() POV we just have a negative dentry -
3471 * all is fine. Let's be bastards - you had / on the end, you've
3472 * been asking for (non-existent) directory. -ENOENT for you.
3474 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3478 if (unlikely(err2)) {
3486 dentry = ERR_PTR(error);
3488 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3490 mnt_drop_write(nd.path.mnt);
3496 struct dentry *kern_path_create(int dfd, const char *pathname,
3497 struct path *path, unsigned int lookup_flags)
3499 struct filename *filename = getname_kernel(pathname);
3502 if (IS_ERR(filename))
3503 return ERR_CAST(filename);
3504 res = filename_create(dfd, filename, path, lookup_flags);
3508 EXPORT_SYMBOL(kern_path_create);
3510 void done_path_create(struct path *path, struct dentry *dentry)
3513 mutex_unlock(&path->dentry->d_inode->i_mutex);
3514 mnt_drop_write(path->mnt);
3517 EXPORT_SYMBOL(done_path_create);
3519 struct dentry *user_path_create(int dfd, const char __user *pathname,
3520 struct path *path, unsigned int lookup_flags)
3522 struct filename *tmp = getname(pathname);
3525 return ERR_CAST(tmp);
3526 res = filename_create(dfd, tmp, path, lookup_flags);
3530 EXPORT_SYMBOL(user_path_create);
3532 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3534 int error = may_create(dir, dentry);
3539 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3542 if (!dir->i_op->mknod)
3545 error = devcgroup_inode_mknod(mode, dev);
3549 error = security_inode_mknod(dir, dentry, mode, dev);
3553 error = dir->i_op->mknod(dir, dentry, mode, dev);
3555 fsnotify_create(dir, dentry);
3558 EXPORT_SYMBOL(vfs_mknod);
3560 static int may_mknod(umode_t mode)
3562 switch (mode & S_IFMT) {
3568 case 0: /* zero mode translates to S_IFREG */
3577 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3580 struct dentry *dentry;
3583 unsigned int lookup_flags = 0;
3585 error = may_mknod(mode);
3589 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3591 return PTR_ERR(dentry);
3593 if (!IS_POSIXACL(path.dentry->d_inode))
3594 mode &= ~current_umask();
3595 error = security_path_mknod(&path, dentry, mode, dev);
3598 switch (mode & S_IFMT) {
3599 case 0: case S_IFREG:
3600 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3602 case S_IFCHR: case S_IFBLK:
3603 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3604 new_decode_dev(dev));
3606 case S_IFIFO: case S_IFSOCK:
3607 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3611 done_path_create(&path, dentry);
3612 if (retry_estale(error, lookup_flags)) {
3613 lookup_flags |= LOOKUP_REVAL;
3619 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3621 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3624 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3626 int error = may_create(dir, dentry);
3627 unsigned max_links = dir->i_sb->s_max_links;
3632 if (!dir->i_op->mkdir)
3635 mode &= (S_IRWXUGO|S_ISVTX);
3636 error = security_inode_mkdir(dir, dentry, mode);
3640 if (max_links && dir->i_nlink >= max_links)
3643 error = dir->i_op->mkdir(dir, dentry, mode);
3645 fsnotify_mkdir(dir, dentry);
3648 EXPORT_SYMBOL(vfs_mkdir);
3650 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3652 struct dentry *dentry;
3655 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3658 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3660 return PTR_ERR(dentry);
3662 if (!IS_POSIXACL(path.dentry->d_inode))
3663 mode &= ~current_umask();
3664 error = security_path_mkdir(&path, dentry, mode);
3666 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3667 done_path_create(&path, dentry);
3668 if (retry_estale(error, lookup_flags)) {
3669 lookup_flags |= LOOKUP_REVAL;
3675 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3677 return sys_mkdirat(AT_FDCWD, pathname, mode);
3681 * The dentry_unhash() helper will try to drop the dentry early: we
3682 * should have a usage count of 1 if we're the only user of this
3683 * dentry, and if that is true (possibly after pruning the dcache),
3684 * then we drop the dentry now.
3686 * A low-level filesystem can, if it choses, legally
3689 * if (!d_unhashed(dentry))
3692 * if it cannot handle the case of removing a directory
3693 * that is still in use by something else..
3695 void dentry_unhash(struct dentry *dentry)
3697 shrink_dcache_parent(dentry);
3698 spin_lock(&dentry->d_lock);
3699 if (dentry->d_lockref.count == 1)
3701 spin_unlock(&dentry->d_lock);
3703 EXPORT_SYMBOL(dentry_unhash);
3705 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3707 int error = may_delete(dir, dentry, 1);
3712 if (!dir->i_op->rmdir)
3716 mutex_lock(&dentry->d_inode->i_mutex);
3719 if (is_local_mountpoint(dentry))
3722 error = security_inode_rmdir(dir, dentry);
3726 shrink_dcache_parent(dentry);
3727 error = dir->i_op->rmdir(dir, dentry);
3731 dentry->d_inode->i_flags |= S_DEAD;
3733 detach_mounts(dentry);
3736 mutex_unlock(&dentry->d_inode->i_mutex);
3742 EXPORT_SYMBOL(vfs_rmdir);
3744 static long do_rmdir(int dfd, const char __user *pathname)
3747 struct filename *name;
3748 struct dentry *dentry;
3752 unsigned int lookup_flags = 0;
3754 name = user_path_parent(dfd, pathname,
3755 &path, &last, &type, lookup_flags);
3757 return PTR_ERR(name);
3771 error = mnt_want_write(path.mnt);
3775 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3776 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3777 error = PTR_ERR(dentry);
3780 if (!dentry->d_inode) {
3784 error = security_path_rmdir(&path, dentry);
3787 error = vfs_rmdir(path.dentry->d_inode, dentry);
3791 mutex_unlock(&path.dentry->d_inode->i_mutex);
3792 mnt_drop_write(path.mnt);
3796 if (retry_estale(error, lookup_flags)) {
3797 lookup_flags |= LOOKUP_REVAL;
3803 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3805 return do_rmdir(AT_FDCWD, pathname);
3809 * vfs_unlink - unlink a filesystem object
3810 * @dir: parent directory
3812 * @delegated_inode: returns victim inode, if the inode is delegated.
3814 * The caller must hold dir->i_mutex.
3816 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3817 * return a reference to the inode in delegated_inode. The caller
3818 * should then break the delegation on that inode and retry. Because
3819 * breaking a delegation may take a long time, the caller should drop
3820 * dir->i_mutex before doing so.
3822 * Alternatively, a caller may pass NULL for delegated_inode. This may
3823 * be appropriate for callers that expect the underlying filesystem not
3824 * to be NFS exported.
3826 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3828 struct inode *target = dentry->d_inode;
3829 int error = may_delete(dir, dentry, 0);
3834 if (!dir->i_op->unlink)
3837 mutex_lock(&target->i_mutex);
3838 if (is_local_mountpoint(dentry))
3841 error = security_inode_unlink(dir, dentry);
3843 error = try_break_deleg(target, delegated_inode);
3846 error = dir->i_op->unlink(dir, dentry);
3849 detach_mounts(dentry);
3854 mutex_unlock(&target->i_mutex);
3856 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3857 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3858 fsnotify_link_count(target);
3864 EXPORT_SYMBOL(vfs_unlink);
3867 * Make sure that the actual truncation of the file will occur outside its
3868 * directory's i_mutex. Truncate can take a long time if there is a lot of
3869 * writeout happening, and we don't want to prevent access to the directory
3870 * while waiting on the I/O.
3872 static long do_unlinkat(int dfd, const char __user *pathname)
3875 struct filename *name;
3876 struct dentry *dentry;
3880 struct inode *inode = NULL;
3881 struct inode *delegated_inode = NULL;
3882 unsigned int lookup_flags = 0;
3884 name = user_path_parent(dfd, pathname,
3885 &path, &last, &type, lookup_flags);
3887 return PTR_ERR(name);
3890 if (type != LAST_NORM)
3893 error = mnt_want_write(path.mnt);
3897 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3898 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3899 error = PTR_ERR(dentry);
3900 if (!IS_ERR(dentry)) {
3901 /* Why not before? Because we want correct error value */
3902 if (last.name[last.len])
3904 inode = dentry->d_inode;
3905 if (d_is_negative(dentry))
3908 error = security_path_unlink(&path, dentry);
3911 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3915 mutex_unlock(&path.dentry->d_inode->i_mutex);
3917 iput(inode); /* truncate the inode here */
3919 if (delegated_inode) {
3920 error = break_deleg_wait(&delegated_inode);
3924 mnt_drop_write(path.mnt);
3928 if (retry_estale(error, lookup_flags)) {
3929 lookup_flags |= LOOKUP_REVAL;
3936 if (d_is_negative(dentry))
3938 else if (d_is_dir(dentry))
3945 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3947 if ((flag & ~AT_REMOVEDIR) != 0)
3950 if (flag & AT_REMOVEDIR)
3951 return do_rmdir(dfd, pathname);
3953 return do_unlinkat(dfd, pathname);
3956 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3958 return do_unlinkat(AT_FDCWD, pathname);
3961 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3963 int error = may_create(dir, dentry);
3968 if (!dir->i_op->symlink)
3971 error = security_inode_symlink(dir, dentry, oldname);
3975 error = dir->i_op->symlink(dir, dentry, oldname);
3977 fsnotify_create(dir, dentry);
3980 EXPORT_SYMBOL(vfs_symlink);
3982 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3983 int, newdfd, const char __user *, newname)
3986 struct filename *from;
3987 struct dentry *dentry;
3989 unsigned int lookup_flags = 0;
3991 from = getname(oldname);
3993 return PTR_ERR(from);
3995 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3996 error = PTR_ERR(dentry);
4000 error = security_path_symlink(&path, dentry, from->name);
4002 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4003 done_path_create(&path, dentry);
4004 if (retry_estale(error, lookup_flags)) {
4005 lookup_flags |= LOOKUP_REVAL;
4013 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4015 return sys_symlinkat(oldname, AT_FDCWD, newname);
4019 * vfs_link - create a new link
4020 * @old_dentry: object to be linked
4022 * @new_dentry: where to create the new link
4023 * @delegated_inode: returns inode needing a delegation break
4025 * The caller must hold dir->i_mutex
4027 * If vfs_link discovers a delegation on the to-be-linked file in need
4028 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4029 * inode in delegated_inode. The caller should then break the delegation
4030 * and retry. Because breaking a delegation may take a long time, the
4031 * caller should drop the i_mutex before doing so.
4033 * Alternatively, a caller may pass NULL for delegated_inode. This may
4034 * be appropriate for callers that expect the underlying filesystem not
4035 * to be NFS exported.
4037 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4039 struct inode *inode = old_dentry->d_inode;
4040 unsigned max_links = dir->i_sb->s_max_links;
4046 error = may_create(dir, new_dentry);
4050 if (dir->i_sb != inode->i_sb)
4054 * A link to an append-only or immutable file cannot be created.
4056 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4058 if (!dir->i_op->link)
4060 if (S_ISDIR(inode->i_mode))
4063 error = security_inode_link(old_dentry, dir, new_dentry);
4067 mutex_lock(&inode->i_mutex);
4068 /* Make sure we don't allow creating hardlink to an unlinked file */
4069 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4071 else if (max_links && inode->i_nlink >= max_links)
4074 error = try_break_deleg(inode, delegated_inode);
4076 error = dir->i_op->link(old_dentry, dir, new_dentry);
4079 if (!error && (inode->i_state & I_LINKABLE)) {
4080 spin_lock(&inode->i_lock);
4081 inode->i_state &= ~I_LINKABLE;
4082 spin_unlock(&inode->i_lock);
4084 mutex_unlock(&inode->i_mutex);
4086 fsnotify_link(dir, inode, new_dentry);
4089 EXPORT_SYMBOL(vfs_link);
4092 * Hardlinks are often used in delicate situations. We avoid
4093 * security-related surprises by not following symlinks on the
4096 * We don't follow them on the oldname either to be compatible
4097 * with linux 2.0, and to avoid hard-linking to directories
4098 * and other special files. --ADM
4100 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4101 int, newdfd, const char __user *, newname, int, flags)
4103 struct dentry *new_dentry;
4104 struct path old_path, new_path;
4105 struct inode *delegated_inode = NULL;
4109 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4112 * To use null names we require CAP_DAC_READ_SEARCH
4113 * This ensures that not everyone will be able to create
4114 * handlink using the passed filedescriptor.
4116 if (flags & AT_EMPTY_PATH) {
4117 if (!capable(CAP_DAC_READ_SEARCH))
4122 if (flags & AT_SYMLINK_FOLLOW)
4123 how |= LOOKUP_FOLLOW;
4125 error = user_path_at(olddfd, oldname, how, &old_path);
4129 new_dentry = user_path_create(newdfd, newname, &new_path,
4130 (how & LOOKUP_REVAL));
4131 error = PTR_ERR(new_dentry);
4132 if (IS_ERR(new_dentry))
4136 if (old_path.mnt != new_path.mnt)
4138 error = may_linkat(&old_path);
4139 if (unlikely(error))
4141 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4144 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4146 done_path_create(&new_path, new_dentry);
4147 if (delegated_inode) {
4148 error = break_deleg_wait(&delegated_inode);
4150 path_put(&old_path);
4154 if (retry_estale(error, how)) {
4155 path_put(&old_path);
4156 how |= LOOKUP_REVAL;
4160 path_put(&old_path);
4165 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4167 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4171 * vfs_rename - rename a filesystem object
4172 * @old_dir: parent of source
4173 * @old_dentry: source
4174 * @new_dir: parent of destination
4175 * @new_dentry: destination
4176 * @delegated_inode: returns an inode needing a delegation break
4177 * @flags: rename flags
4179 * The caller must hold multiple mutexes--see lock_rename()).
4181 * If vfs_rename discovers a delegation in need of breaking at either
4182 * the source or destination, it will return -EWOULDBLOCK and return a
4183 * reference to the inode in delegated_inode. The caller should then
4184 * break the delegation and retry. Because breaking a delegation may
4185 * take a long time, the caller should drop all locks before doing
4188 * Alternatively, a caller may pass NULL for delegated_inode. This may
4189 * be appropriate for callers that expect the underlying filesystem not
4190 * to be NFS exported.
4192 * The worst of all namespace operations - renaming directory. "Perverted"
4193 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4195 * a) we can get into loop creation.
4196 * b) race potential - two innocent renames can create a loop together.
4197 * That's where 4.4 screws up. Current fix: serialization on
4198 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4200 * c) we have to lock _four_ objects - parents and victim (if it exists),
4201 * and source (if it is not a directory).
4202 * And that - after we got ->i_mutex on parents (until then we don't know
4203 * whether the target exists). Solution: try to be smart with locking
4204 * order for inodes. We rely on the fact that tree topology may change
4205 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4206 * move will be locked. Thus we can rank directories by the tree
4207 * (ancestors first) and rank all non-directories after them.
4208 * That works since everybody except rename does "lock parent, lookup,
4209 * lock child" and rename is under ->s_vfs_rename_mutex.
4210 * HOWEVER, it relies on the assumption that any object with ->lookup()
4211 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4212 * we'd better make sure that there's no link(2) for them.
4213 * d) conversion from fhandle to dentry may come in the wrong moment - when
4214 * we are removing the target. Solution: we will have to grab ->i_mutex
4215 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4216 * ->i_mutex on parents, which works but leads to some truly excessive
4219 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4220 struct inode *new_dir, struct dentry *new_dentry,
4221 struct inode **delegated_inode, unsigned int flags)
4224 bool is_dir = d_is_dir(old_dentry);
4225 const unsigned char *old_name;
4226 struct inode *source = old_dentry->d_inode;
4227 struct inode *target = new_dentry->d_inode;
4228 bool new_is_dir = false;
4229 unsigned max_links = new_dir->i_sb->s_max_links;
4231 if (source == target)
4234 error = may_delete(old_dir, old_dentry, is_dir);
4239 error = may_create(new_dir, new_dentry);
4241 new_is_dir = d_is_dir(new_dentry);
4243 if (!(flags & RENAME_EXCHANGE))
4244 error = may_delete(new_dir, new_dentry, is_dir);
4246 error = may_delete(new_dir, new_dentry, new_is_dir);
4251 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4254 if (flags && !old_dir->i_op->rename2)
4258 * If we are going to change the parent - check write permissions,
4259 * we'll need to flip '..'.
4261 if (new_dir != old_dir) {
4263 error = inode_permission(source, MAY_WRITE);
4267 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4268 error = inode_permission(target, MAY_WRITE);
4274 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4279 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4281 if (!is_dir || (flags & RENAME_EXCHANGE))
4282 lock_two_nondirectories(source, target);
4284 mutex_lock(&target->i_mutex);
4287 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4290 if (max_links && new_dir != old_dir) {
4292 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4294 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4295 old_dir->i_nlink >= max_links)
4298 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4299 shrink_dcache_parent(new_dentry);
4301 error = try_break_deleg(source, delegated_inode);
4305 if (target && !new_is_dir) {
4306 error = try_break_deleg(target, delegated_inode);
4310 if (!old_dir->i_op->rename2) {
4311 error = old_dir->i_op->rename(old_dir, old_dentry,
4312 new_dir, new_dentry);
4314 WARN_ON(old_dir->i_op->rename != NULL);
4315 error = old_dir->i_op->rename2(old_dir, old_dentry,
4316 new_dir, new_dentry, flags);
4321 if (!(flags & RENAME_EXCHANGE) && target) {
4323 target->i_flags |= S_DEAD;
4324 dont_mount(new_dentry);
4325 detach_mounts(new_dentry);
4327 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4328 if (!(flags & RENAME_EXCHANGE))
4329 d_move(old_dentry, new_dentry);
4331 d_exchange(old_dentry, new_dentry);
4334 if (!is_dir || (flags & RENAME_EXCHANGE))
4335 unlock_two_nondirectories(source, target);
4337 mutex_unlock(&target->i_mutex);
4340 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4341 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4342 if (flags & RENAME_EXCHANGE) {
4343 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4344 new_is_dir, NULL, new_dentry);
4347 fsnotify_oldname_free(old_name);
4351 EXPORT_SYMBOL(vfs_rename);
4353 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4354 int, newdfd, const char __user *, newname, unsigned int, flags)
4356 struct dentry *old_dentry, *new_dentry;
4357 struct dentry *trap;
4358 struct path old_path, new_path;
4359 struct qstr old_last, new_last;
4360 int old_type, new_type;
4361 struct inode *delegated_inode = NULL;
4362 struct filename *from;
4363 struct filename *to;
4364 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4365 bool should_retry = false;
4368 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4371 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4372 (flags & RENAME_EXCHANGE))
4375 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4378 if (flags & RENAME_EXCHANGE)
4382 from = user_path_parent(olddfd, oldname,
4383 &old_path, &old_last, &old_type, lookup_flags);
4385 error = PTR_ERR(from);
4389 to = user_path_parent(newdfd, newname,
4390 &new_path, &new_last, &new_type, lookup_flags);
4392 error = PTR_ERR(to);
4397 if (old_path.mnt != new_path.mnt)
4401 if (old_type != LAST_NORM)
4404 if (flags & RENAME_NOREPLACE)
4406 if (new_type != LAST_NORM)
4409 error = mnt_want_write(old_path.mnt);
4414 trap = lock_rename(new_path.dentry, old_path.dentry);
4416 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4417 error = PTR_ERR(old_dentry);
4418 if (IS_ERR(old_dentry))
4420 /* source must exist */
4422 if (d_is_negative(old_dentry))
4424 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4425 error = PTR_ERR(new_dentry);
4426 if (IS_ERR(new_dentry))
4429 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4431 if (flags & RENAME_EXCHANGE) {
4433 if (d_is_negative(new_dentry))
4436 if (!d_is_dir(new_dentry)) {
4438 if (new_last.name[new_last.len])
4442 /* unless the source is a directory trailing slashes give -ENOTDIR */
4443 if (!d_is_dir(old_dentry)) {
4445 if (old_last.name[old_last.len])
4447 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4450 /* source should not be ancestor of target */
4452 if (old_dentry == trap)
4454 /* target should not be an ancestor of source */
4455 if (!(flags & RENAME_EXCHANGE))
4457 if (new_dentry == trap)
4460 error = security_path_rename(&old_path, old_dentry,
4461 &new_path, new_dentry, flags);
4464 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4465 new_path.dentry->d_inode, new_dentry,
4466 &delegated_inode, flags);
4472 unlock_rename(new_path.dentry, old_path.dentry);
4473 if (delegated_inode) {
4474 error = break_deleg_wait(&delegated_inode);
4478 mnt_drop_write(old_path.mnt);
4480 if (retry_estale(error, lookup_flags))
4481 should_retry = true;
4482 path_put(&new_path);
4485 path_put(&old_path);
4488 should_retry = false;
4489 lookup_flags |= LOOKUP_REVAL;
4496 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4497 int, newdfd, const char __user *, newname)
4499 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4502 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4504 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4507 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4509 int error = may_create(dir, dentry);
4513 if (!dir->i_op->mknod)
4516 return dir->i_op->mknod(dir, dentry,
4517 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4519 EXPORT_SYMBOL(vfs_whiteout);
4521 int readlink_copy(char __user *buffer, int buflen, const char *link)
4523 int len = PTR_ERR(link);
4528 if (len > (unsigned) buflen)
4530 if (copy_to_user(buffer, link, len))
4535 EXPORT_SYMBOL(readlink_copy);
4538 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4539 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4540 * using) it for any given inode is up to filesystem.
4542 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4545 struct inode *inode = d_inode(dentry);
4546 const char *link = inode->i_link;
4550 link = inode->i_op->follow_link(dentry, &cookie);
4552 return PTR_ERR(link);
4554 res = readlink_copy(buffer, buflen, link);
4555 if (inode->i_op->put_link)
4556 inode->i_op->put_link(inode, cookie);
4559 EXPORT_SYMBOL(generic_readlink);
4561 /* get the link contents into pagecache */
4562 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4566 struct address_space *mapping = dentry->d_inode->i_mapping;
4567 page = read_mapping_page(mapping, 0, NULL);
4572 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4576 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4578 struct page *page = NULL;
4579 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4582 page_cache_release(page);
4586 EXPORT_SYMBOL(page_readlink);
4588 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4590 struct page *page = NULL;
4591 char *res = page_getlink(dentry, &page);
4596 EXPORT_SYMBOL(page_follow_link_light);
4598 void page_put_link(struct inode *unused, void *cookie)
4600 struct page *page = cookie;
4602 page_cache_release(page);
4604 EXPORT_SYMBOL(page_put_link);
4607 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4609 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4611 struct address_space *mapping = inode->i_mapping;
4616 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4618 flags |= AOP_FLAG_NOFS;
4621 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4622 flags, &page, &fsdata);
4626 kaddr = kmap_atomic(page);
4627 memcpy(kaddr, symname, len-1);
4628 kunmap_atomic(kaddr);
4630 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4637 mark_inode_dirty(inode);
4642 EXPORT_SYMBOL(__page_symlink);
4644 int page_symlink(struct inode *inode, const char *symname, int len)
4646 return __page_symlink(inode, symname, len,
4647 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4649 EXPORT_SYMBOL(page_symlink);
4651 const struct inode_operations page_symlink_inode_operations = {
4652 .readlink = generic_readlink,
4653 .follow_link = page_follow_link_light,
4654 .put_link = page_put_link,
4656 EXPORT_SYMBOL(page_symlink_inode_operations);