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/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user *filename, char *page)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 char * getname(const char __user * filename)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
175 umode_t mode = inode->i_mode;
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179 if (current_fsuid() == inode->i_uid)
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
188 if (in_group_p(inode->i_gid))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask & ~mode) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
222 * Do the basic POSIX ACL permission checks.
224 ret = acl_permission_check(inode, mask, flags, check_acl);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask & MAY_EXEC) || execute_ok(inode))
233 if (capable(CAP_DAC_OVERRIDE))
237 * Searching includes executable on directories, else just read.
239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode *inode, int mask)
261 if (mask & MAY_WRITE) {
262 umode_t mode = inode->i_mode;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode))
278 if (inode->i_op->permission)
279 retval = inode->i_op->permission(inode, mask, 0);
281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
287 retval = devcgroup_inode_permission(inode, mask);
291 return security_inode_permission(inode, mask);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file *file, int mask)
308 return inode_permission(file->f_path.dentry->d_inode, mask);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode * inode)
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
342 int deny_write_access(struct file * file)
344 struct inode *inode = file->f_path.dentry->d_inode;
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path *path)
368 EXPORT_SYMBOL(path_get);
371 * path_get_long - get a long reference to a path
372 * @path: path to get the reference to
374 * Given a path increment the reference count to the dentry and the vfsmount.
376 void path_get_long(struct path *path)
378 mntget_long(path->mnt);
383 * path_put - put a reference to a path
384 * @path: path to put the reference to
386 * Given a path decrement the reference count to the dentry and the vfsmount.
388 void path_put(struct path *path)
393 EXPORT_SYMBOL(path_put);
396 * path_put_long - put a long reference to a path
397 * @path: path to put the reference to
399 * Given a path decrement the reference count to the dentry and the vfsmount.
401 void path_put_long(struct path *path)
404 mntput_long(path->mnt);
408 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
409 * @nd: nameidata pathwalk data to drop
410 * Returns: 0 on success, -ECHILD on failure
412 * Path walking has 2 modes, rcu-walk and ref-walk (see
413 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
414 * to drop out of rcu-walk mode and take normal reference counts on dentries
415 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
416 * refcounts at the last known good point before rcu-walk got stuck, so
417 * ref-walk may continue from there. If this is not successful (eg. a seqcount
418 * has changed), then failure is returned and path walk restarts from the
419 * beginning in ref-walk mode.
421 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
422 * ref-walk. Must be called from rcu-walk context.
424 static int nameidata_drop_rcu(struct nameidata *nd)
426 struct fs_struct *fs = current->fs;
427 struct dentry *dentry = nd->path.dentry;
429 BUG_ON(!(nd->flags & LOOKUP_RCU));
431 spin_lock(&fs->lock);
432 if (nd->root.mnt != fs->root.mnt ||
433 nd->root.dentry != fs->root.dentry)
436 spin_lock(&dentry->d_lock);
437 if (!__d_rcu_to_refcount(dentry, nd->seq))
439 BUG_ON(nd->inode != dentry->d_inode);
440 spin_unlock(&dentry->d_lock);
443 spin_unlock(&fs->lock);
445 mntget(nd->path.mnt);
448 br_read_unlock(vfsmount_lock);
449 nd->flags &= ~LOOKUP_RCU;
452 spin_unlock(&dentry->d_lock);
455 spin_unlock(&fs->lock);
459 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
460 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
462 if (nd->flags & LOOKUP_RCU)
463 return nameidata_drop_rcu(nd);
468 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
469 * @nd: nameidata pathwalk data to drop
470 * @dentry: dentry to drop
471 * Returns: 0 on success, -ECHILD on failure
473 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
474 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
475 * @nd. Must be called from rcu-walk context.
477 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
479 struct fs_struct *fs = current->fs;
480 struct dentry *parent = nd->path.dentry;
483 * It can be possible to revalidate the dentry that we started
484 * the path walk with. force_reval_path may also revalidate the
485 * dentry already committed to the nameidata.
487 if (unlikely(parent == dentry))
488 return nameidata_drop_rcu(nd);
490 BUG_ON(!(nd->flags & LOOKUP_RCU));
492 spin_lock(&fs->lock);
493 if (nd->root.mnt != fs->root.mnt ||
494 nd->root.dentry != fs->root.dentry)
497 spin_lock(&parent->d_lock);
498 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
499 if (!__d_rcu_to_refcount(dentry, nd->seq))
502 * If the sequence check on the child dentry passed, then the child has
503 * not been removed from its parent. This means the parent dentry must
504 * be valid and able to take a reference at this point.
506 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
507 BUG_ON(!parent->d_count);
509 spin_unlock(&dentry->d_lock);
510 spin_unlock(&parent->d_lock);
513 spin_unlock(&fs->lock);
515 mntget(nd->path.mnt);
518 br_read_unlock(vfsmount_lock);
519 nd->flags &= ~LOOKUP_RCU;
522 spin_unlock(&dentry->d_lock);
523 spin_unlock(&parent->d_lock);
526 spin_unlock(&fs->lock);
530 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
531 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
533 if (nd->flags & LOOKUP_RCU)
534 return nameidata_dentry_drop_rcu(nd, dentry);
539 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
540 * @nd: nameidata pathwalk data to drop
541 * Returns: 0 on success, -ECHILD on failure
543 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
544 * nd->path should be the final element of the lookup, so nd->root is discarded.
545 * Must be called from rcu-walk context.
547 static int nameidata_drop_rcu_last(struct nameidata *nd)
549 struct dentry *dentry = nd->path.dentry;
551 BUG_ON(!(nd->flags & LOOKUP_RCU));
552 nd->flags &= ~LOOKUP_RCU;
554 spin_lock(&dentry->d_lock);
555 if (!__d_rcu_to_refcount(dentry, nd->seq))
557 BUG_ON(nd->inode != dentry->d_inode);
558 spin_unlock(&dentry->d_lock);
560 mntget(nd->path.mnt);
563 br_read_unlock(vfsmount_lock);
568 spin_unlock(&dentry->d_lock);
570 br_read_unlock(vfsmount_lock);
574 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
575 static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
577 if (likely(nd->flags & LOOKUP_RCU))
578 return nameidata_drop_rcu_last(nd);
583 * release_open_intent - free up open intent resources
584 * @nd: pointer to nameidata
586 void release_open_intent(struct nameidata *nd)
588 if (nd->intent.open.file->f_path.dentry == NULL)
589 put_filp(nd->intent.open.file);
591 fput(nd->intent.open.file);
595 * Call d_revalidate and handle filesystems that request rcu-walk
596 * to be dropped. This may be called and return in rcu-walk mode,
597 * regardless of success or error. If -ECHILD is returned, the caller
598 * must return -ECHILD back up the path walk stack so path walk may
599 * be restarted in ref-walk mode.
601 static int d_revalidate(struct dentry *dentry, struct nameidata *nd)
605 status = dentry->d_op->d_revalidate(dentry, nd);
606 if (status == -ECHILD) {
607 if (nameidata_dentry_drop_rcu(nd, dentry))
609 status = dentry->d_op->d_revalidate(dentry, nd);
615 static inline struct dentry *
616 do_revalidate(struct dentry *dentry, struct nameidata *nd)
620 status = d_revalidate(dentry, nd);
621 if (unlikely(status <= 0)) {
623 * The dentry failed validation.
624 * If d_revalidate returned 0 attempt to invalidate
625 * the dentry otherwise d_revalidate is asking us
626 * to return a fail status.
629 /* If we're in rcu-walk, we don't have a ref */
630 if (!(nd->flags & LOOKUP_RCU))
632 dentry = ERR_PTR(status);
635 /* Don't d_invalidate in rcu-walk mode */
636 if (nameidata_dentry_drop_rcu_maybe(nd, dentry))
637 return ERR_PTR(-ECHILD);
638 if (!d_invalidate(dentry)) {
647 static inline int need_reval_dot(struct dentry *dentry)
649 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
652 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
659 * force_reval_path - force revalidation of a dentry
661 * In some situations the path walking code will trust dentries without
662 * revalidating them. This causes problems for filesystems that depend on
663 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
664 * (which indicates that it's possible for the dentry to go stale), force
665 * a d_revalidate call before proceeding.
667 * Returns 0 if the revalidation was successful. If the revalidation fails,
668 * either return the error returned by d_revalidate or -ESTALE if the
669 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
670 * invalidate the dentry. It's up to the caller to handle putting references
671 * to the path if necessary.
674 force_reval_path(struct path *path, struct nameidata *nd)
677 struct dentry *dentry = path->dentry;
680 * only check on filesystems where it's possible for the dentry to
683 if (!need_reval_dot(dentry))
686 status = d_revalidate(dentry, nd);
691 /* Don't d_invalidate in rcu-walk mode */
692 if (nameidata_drop_rcu(nd))
694 d_invalidate(dentry);
701 * Short-cut version of permission(), for calling on directories
702 * during pathname resolution. Combines parts of permission()
703 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
705 * If appropriate, check DAC only. If not appropriate, or
706 * short-cut DAC fails, then call ->permission() to do more
707 * complete permission check.
709 static inline int exec_permission(struct inode *inode, unsigned int flags)
713 if (inode->i_op->permission) {
714 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
716 ret = acl_permission_check(inode, MAY_EXEC, flags,
717 inode->i_op->check_acl);
724 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
729 return security_inode_exec_permission(inode, flags);
732 static __always_inline void set_root(struct nameidata *nd)
735 get_fs_root(current->fs, &nd->root);
738 static int link_path_walk(const char *, struct nameidata *);
740 static __always_inline void set_root_rcu(struct nameidata *nd)
743 struct fs_struct *fs = current->fs;
747 seq = read_seqcount_begin(&fs->seq);
749 } while (read_seqcount_retry(&fs->seq, seq));
753 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
766 nd->inode = nd->path.dentry->d_inode;
768 ret = link_path_walk(link, nd);
772 return PTR_ERR(link);
775 static void path_put_conditional(struct path *path, struct nameidata *nd)
778 if (path->mnt != nd->path.mnt)
782 static inline void path_to_nameidata(const struct path *path,
783 struct nameidata *nd)
785 if (!(nd->flags & LOOKUP_RCU)) {
786 dput(nd->path.dentry);
787 if (nd->path.mnt != path->mnt)
788 mntput(nd->path.mnt);
790 nd->path.mnt = path->mnt;
791 nd->path.dentry = path->dentry;
794 static __always_inline int
795 __do_follow_link(const struct path *link, struct nameidata *nd, void **p)
798 struct dentry *dentry = link->dentry;
800 touch_atime(link->mnt, dentry);
801 nd_set_link(nd, NULL);
803 if (link->mnt == nd->path.mnt)
806 nd->last_type = LAST_BIND;
807 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
810 char *s = nd_get_link(nd);
813 error = __vfs_follow_link(nd, s);
814 else if (nd->last_type == LAST_BIND) {
815 error = force_reval_path(&nd->path, nd);
824 * This limits recursive symlink follows to 8, while
825 * limiting consecutive symlinks to 40.
827 * Without that kind of total limit, nasty chains of consecutive
828 * symlinks can cause almost arbitrarily long lookups.
830 static inline int do_follow_link(struct path *path, struct nameidata *nd)
834 if (current->link_count >= MAX_NESTED_LINKS)
836 if (current->total_link_count >= 40)
838 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
840 err = security_inode_follow_link(path->dentry, nd);
843 current->link_count++;
844 current->total_link_count++;
846 err = __do_follow_link(path, nd, &cookie);
847 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
848 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
850 current->link_count--;
854 path_put_conditional(path, nd);
859 static int follow_up_rcu(struct path *path)
861 struct vfsmount *parent;
862 struct dentry *mountpoint;
864 parent = path->mnt->mnt_parent;
865 if (parent == path->mnt)
867 mountpoint = path->mnt->mnt_mountpoint;
868 path->dentry = mountpoint;
873 int follow_up(struct path *path)
875 struct vfsmount *parent;
876 struct dentry *mountpoint;
878 br_read_lock(vfsmount_lock);
879 parent = path->mnt->mnt_parent;
880 if (parent == path->mnt) {
881 br_read_unlock(vfsmount_lock);
885 mountpoint = dget(path->mnt->mnt_mountpoint);
886 br_read_unlock(vfsmount_lock);
888 path->dentry = mountpoint;
895 * Perform an automount
896 * - return -EISDIR to tell follow_managed() to stop and return the path we
899 static int follow_automount(struct path *path, unsigned flags,
902 struct vfsmount *mnt;
904 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
907 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
908 * and this is the terminal part of the path.
910 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
911 return -EISDIR; /* we actually want to stop here */
913 /* We want to mount if someone is trying to open/create a file of any
914 * type under the mountpoint, wants to traverse through the mountpoint
915 * or wants to open the mounted directory.
917 * We don't want to mount if someone's just doing a stat and they've
918 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
919 * appended a '/' to the name.
921 if (!(flags & LOOKUP_FOLLOW) &&
922 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
923 LOOKUP_OPEN | LOOKUP_CREATE)))
926 current->total_link_count++;
927 if (current->total_link_count >= 40)
930 mnt = path->dentry->d_op->d_automount(path);
933 * The filesystem is allowed to return -EISDIR here to indicate
934 * it doesn't want to automount. For instance, autofs would do
935 * this so that its userspace daemon can mount on this dentry.
937 * However, we can only permit this if it's a terminal point in
938 * the path being looked up; if it wasn't then the remainder of
939 * the path is inaccessible and we should say so.
941 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
945 if (!mnt) /* mount collision */
948 if (mnt->mnt_sb == path->mnt->mnt_sb &&
949 mnt->mnt_root == path->dentry) {
958 path->dentry = dget(mnt->mnt_root);
964 * Handle a dentry that is managed in some way.
965 * - Flagged for transit management (autofs)
966 * - Flagged as mountpoint
967 * - Flagged as automount point
969 * This may only be called in refwalk mode.
971 * Serialization is taken care of in namespace.c
973 static int follow_managed(struct path *path, unsigned flags)
976 bool need_mntput = false;
979 /* Given that we're not holding a lock here, we retain the value in a
980 * local variable for each dentry as we look at it so that we don't see
981 * the components of that value change under us */
982 while (managed = ACCESS_ONCE(path->dentry->d_flags),
983 managed &= DCACHE_MANAGED_DENTRY,
984 unlikely(managed != 0)) {
985 /* Allow the filesystem to manage the transit without i_mutex
987 if (managed & DCACHE_MANAGE_TRANSIT) {
988 BUG_ON(!path->dentry->d_op);
989 BUG_ON(!path->dentry->d_op->d_manage);
990 ret = path->dentry->d_op->d_manage(path->dentry, false);
992 return ret == -EISDIR ? 0 : ret;
995 /* Transit to a mounted filesystem. */
996 if (managed & DCACHE_MOUNTED) {
997 struct vfsmount *mounted = lookup_mnt(path);
1002 path->mnt = mounted;
1003 path->dentry = dget(mounted->mnt_root);
1008 /* Something is mounted on this dentry in another
1009 * namespace and/or whatever was mounted there in this
1010 * namespace got unmounted before we managed to get the
1014 /* Handle an automount point */
1015 if (managed & DCACHE_NEED_AUTOMOUNT) {
1016 ret = follow_automount(path, flags, &need_mntput);
1018 return ret == -EISDIR ? 0 : ret;
1022 /* We didn't change the current path point */
1028 int follow_down_one(struct path *path)
1030 struct vfsmount *mounted;
1032 mounted = lookup_mnt(path);
1036 path->mnt = mounted;
1037 path->dentry = dget(mounted->mnt_root);
1044 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1045 * meet a managed dentry and we're not walking to "..". True is returned to
1046 * continue, false to abort.
1048 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1049 struct inode **inode, bool reverse_transit)
1051 unsigned abort_mask =
1052 reverse_transit ? 0 : DCACHE_MANAGE_TRANSIT;
1054 while (d_mountpoint(path->dentry)) {
1055 struct vfsmount *mounted;
1056 if (path->dentry->d_flags & abort_mask)
1058 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1061 path->mnt = mounted;
1062 path->dentry = mounted->mnt_root;
1063 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1064 *inode = path->dentry->d_inode;
1067 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1068 return reverse_transit;
1072 static int follow_dotdot_rcu(struct nameidata *nd)
1074 struct inode *inode = nd->inode;
1079 if (nd->path.dentry == nd->root.dentry &&
1080 nd->path.mnt == nd->root.mnt) {
1083 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1084 struct dentry *old = nd->path.dentry;
1085 struct dentry *parent = old->d_parent;
1088 seq = read_seqcount_begin(&parent->d_seq);
1089 if (read_seqcount_retry(&old->d_seq, nd->seq))
1091 inode = parent->d_inode;
1092 nd->path.dentry = parent;
1096 if (!follow_up_rcu(&nd->path))
1098 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1099 inode = nd->path.dentry->d_inode;
1101 __follow_mount_rcu(nd, &nd->path, &inode, true);
1108 * Follow down to the covering mount currently visible to userspace. At each
1109 * point, the filesystem owning that dentry may be queried as to whether the
1110 * caller is permitted to proceed or not.
1112 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1115 int follow_down(struct path *path, bool mounting_here)
1120 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1121 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1122 /* Allow the filesystem to manage the transit without i_mutex
1125 * We indicate to the filesystem if someone is trying to mount
1126 * something here. This gives autofs the chance to deny anyone
1127 * other than its daemon the right to mount on its
1130 * The filesystem may sleep at this point.
1132 if (managed & DCACHE_MANAGE_TRANSIT) {
1133 BUG_ON(!path->dentry->d_op);
1134 BUG_ON(!path->dentry->d_op->d_manage);
1135 ret = path->dentry->d_op->d_manage(path->dentry, mounting_here);
1137 return ret == -EISDIR ? 0 : ret;
1140 /* Transit to a mounted filesystem. */
1141 if (managed & DCACHE_MOUNTED) {
1142 struct vfsmount *mounted = lookup_mnt(path);
1147 path->mnt = mounted;
1148 path->dentry = dget(mounted->mnt_root);
1152 /* Don't handle automount points here */
1159 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1161 static void follow_mount(struct path *path)
1163 while (d_mountpoint(path->dentry)) {
1164 struct vfsmount *mounted = lookup_mnt(path);
1169 path->mnt = mounted;
1170 path->dentry = dget(mounted->mnt_root);
1174 static void follow_dotdot(struct nameidata *nd)
1179 struct dentry *old = nd->path.dentry;
1181 if (nd->path.dentry == nd->root.dentry &&
1182 nd->path.mnt == nd->root.mnt) {
1185 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1186 /* rare case of legitimate dget_parent()... */
1187 nd->path.dentry = dget_parent(nd->path.dentry);
1191 if (!follow_up(&nd->path))
1194 follow_mount(&nd->path);
1195 nd->inode = nd->path.dentry->d_inode;
1199 * Allocate a dentry with name and parent, and perform a parent
1200 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1201 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1202 * have verified that no child exists while under i_mutex.
1204 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1205 struct qstr *name, struct nameidata *nd)
1207 struct inode *inode = parent->d_inode;
1208 struct dentry *dentry;
1211 /* Don't create child dentry for a dead directory. */
1212 if (unlikely(IS_DEADDIR(inode)))
1213 return ERR_PTR(-ENOENT);
1215 dentry = d_alloc(parent, name);
1216 if (unlikely(!dentry))
1217 return ERR_PTR(-ENOMEM);
1219 old = inode->i_op->lookup(inode, dentry, nd);
1220 if (unlikely(old)) {
1228 * It's more convoluted than I'd like it to be, but... it's still fairly
1229 * small and for now I'd prefer to have fast path as straight as possible.
1230 * It _is_ time-critical.
1232 static int do_lookup(struct nameidata *nd, struct qstr *name,
1233 struct path *path, struct inode **inode)
1235 struct vfsmount *mnt = nd->path.mnt;
1236 struct dentry *dentry, *parent = nd->path.dentry;
1241 * See if the low-level filesystem might want
1242 * to use its own hash..
1244 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1245 err = parent->d_op->d_hash(parent, nd->inode, name);
1251 * Rename seqlock is not required here because in the off chance
1252 * of a false negative due to a concurrent rename, we're going to
1253 * do the non-racy lookup, below.
1255 if (nd->flags & LOOKUP_RCU) {
1259 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1261 if (nameidata_drop_rcu(nd))
1265 /* Memory barrier in read_seqcount_begin of child is enough */
1266 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1270 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
1271 goto need_revalidate;
1274 path->dentry = dentry;
1275 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1277 if (nameidata_drop_rcu(nd))
1281 dentry = __d_lookup(parent, name);
1285 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
1286 goto need_revalidate;
1289 path->dentry = dentry;
1290 err = follow_managed(path, nd->flags);
1291 if (unlikely(err < 0))
1293 *inode = path->dentry->d_inode;
1297 dir = parent->d_inode;
1298 BUG_ON(nd->inode != dir);
1300 mutex_lock(&dir->i_mutex);
1302 * First re-do the cached lookup just in case it was created
1303 * while we waited for the directory semaphore, or the first
1304 * lookup failed due to an unrelated rename.
1306 * This could use version numbering or similar to avoid unnecessary
1307 * cache lookups, but then we'd have to do the first lookup in the
1308 * non-racy way. However in the common case here, everything should
1309 * be hot in cache, so would it be a big win?
1311 dentry = d_lookup(parent, name);
1312 if (likely(!dentry)) {
1313 dentry = d_alloc_and_lookup(parent, name, nd);
1314 mutex_unlock(&dir->i_mutex);
1320 * Uhhuh! Nasty case: the cache was re-populated while
1321 * we waited on the semaphore. Need to revalidate.
1323 mutex_unlock(&dir->i_mutex);
1327 dentry = do_revalidate(dentry, nd);
1332 if (nd->flags & LOOKUP_RCU)
1337 return PTR_ERR(dentry);
1342 * This is the basic name resolution function, turning a pathname into
1343 * the final dentry. We expect 'base' to be positive and a directory.
1345 * Returns 0 and nd will have valid dentry and mnt on success.
1346 * Returns error and drops reference to input namei data on failure.
1348 static int link_path_walk(const char *name, struct nameidata *nd)
1352 unsigned int lookup_flags = nd->flags;
1360 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1362 /* At this point we know we have a real path component. */
1364 struct inode *inode;
1369 nd->flags |= LOOKUP_CONTINUE;
1370 if (nd->flags & LOOKUP_RCU) {
1371 err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1372 if (err == -ECHILD) {
1373 if (nameidata_drop_rcu(nd))
1379 err = exec_permission(nd->inode, 0);
1385 c = *(const unsigned char *)name;
1387 hash = init_name_hash();
1390 hash = partial_name_hash(c, hash);
1391 c = *(const unsigned char *)name;
1392 } while (c && (c != '/'));
1393 this.len = name - (const char *) this.name;
1394 this.hash = end_name_hash(hash);
1396 /* remove trailing slashes? */
1398 goto last_component;
1399 while (*++name == '/');
1401 goto last_with_slashes;
1404 * "." and ".." are special - ".." especially so because it has
1405 * to be able to know about the current root directory and
1406 * parent relationships.
1408 if (this.name[0] == '.') switch (this.len) {
1412 if (this.name[1] != '.')
1414 if (nd->flags & LOOKUP_RCU) {
1415 if (follow_dotdot_rcu(nd))
1423 /* This does the actual lookups.. */
1424 err = do_lookup(nd, &this, &next, &inode);
1431 if (inode->i_op->follow_link) {
1432 /* We commonly drop rcu-walk here */
1433 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1435 BUG_ON(inode != next.dentry->d_inode);
1436 err = do_follow_link(&next, nd);
1439 nd->inode = nd->path.dentry->d_inode;
1444 path_to_nameidata(&next, nd);
1448 if (!nd->inode->i_op->lookup)
1451 /* here ends the main loop */
1454 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1456 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1457 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1458 if (lookup_flags & LOOKUP_PARENT)
1460 if (this.name[0] == '.') switch (this.len) {
1464 if (this.name[1] != '.')
1466 if (nd->flags & LOOKUP_RCU) {
1467 if (follow_dotdot_rcu(nd))
1475 err = do_lookup(nd, &this, &next, &inode);
1478 if (inode && unlikely(inode->i_op->follow_link) &&
1479 (lookup_flags & LOOKUP_FOLLOW)) {
1480 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1482 BUG_ON(inode != next.dentry->d_inode);
1483 err = do_follow_link(&next, nd);
1486 nd->inode = nd->path.dentry->d_inode;
1488 path_to_nameidata(&next, nd);
1494 if (lookup_flags & LOOKUP_DIRECTORY) {
1496 if (!nd->inode->i_op->lookup)
1502 nd->last_type = LAST_NORM;
1503 if (this.name[0] != '.')
1506 nd->last_type = LAST_DOT;
1507 else if (this.len == 2 && this.name[1] == '.')
1508 nd->last_type = LAST_DOTDOT;
1513 * We bypassed the ordinary revalidation routines.
1514 * We may need to check the cached dentry for staleness.
1516 if (need_reval_dot(nd->path.dentry)) {
1517 /* Note: we do not d_invalidate() */
1518 err = d_revalidate(nd->path.dentry, nd);
1525 if (nameidata_drop_rcu_last_maybe(nd))
1529 if (!(nd->flags & LOOKUP_RCU))
1530 path_put_conditional(&next, nd);
1533 if (!(nd->flags & LOOKUP_RCU))
1534 path_put(&nd->path);
1539 static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1541 current->total_link_count = 0;
1543 return link_path_walk(name, nd);
1546 static inline int path_walk_simple(const char *name, struct nameidata *nd)
1548 current->total_link_count = 0;
1550 return link_path_walk(name, nd);
1553 static int path_walk(const char *name, struct nameidata *nd)
1555 struct path save = nd->path;
1558 current->total_link_count = 0;
1560 /* make sure the stuff we saved doesn't go away */
1563 result = link_path_walk(name, nd);
1564 if (result == -ESTALE) {
1565 /* nd->path had been dropped */
1566 current->total_link_count = 0;
1568 path_get(&nd->path);
1569 nd->flags |= LOOKUP_REVAL;
1570 result = link_path_walk(name, nd);
1578 static void path_finish_rcu(struct nameidata *nd)
1580 if (nd->flags & LOOKUP_RCU) {
1581 /* RCU dangling. Cancel it. */
1582 nd->flags &= ~LOOKUP_RCU;
1583 nd->root.mnt = NULL;
1585 br_read_unlock(vfsmount_lock);
1591 static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1597 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1598 nd->flags = flags | LOOKUP_RCU;
1600 nd->root.mnt = NULL;
1604 struct fs_struct *fs = current->fs;
1607 br_read_lock(vfsmount_lock);
1611 seq = read_seqcount_begin(&fs->seq);
1612 nd->root = fs->root;
1613 nd->path = nd->root;
1614 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1615 } while (read_seqcount_retry(&fs->seq, seq));
1617 } else if (dfd == AT_FDCWD) {
1618 struct fs_struct *fs = current->fs;
1621 br_read_lock(vfsmount_lock);
1625 seq = read_seqcount_begin(&fs->seq);
1627 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1628 } while (read_seqcount_retry(&fs->seq, seq));
1631 struct dentry *dentry;
1633 file = fget_light(dfd, &fput_needed);
1638 dentry = file->f_path.dentry;
1641 if (!S_ISDIR(dentry->d_inode->i_mode))
1644 retval = file_permission(file, MAY_EXEC);
1648 nd->path = file->f_path;
1652 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1653 br_read_lock(vfsmount_lock);
1656 nd->inode = nd->path.dentry->d_inode;
1660 fput_light(file, fput_needed);
1665 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1671 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1674 nd->root.mnt = NULL;
1678 nd->path = nd->root;
1679 path_get(&nd->root);
1680 } else if (dfd == AT_FDCWD) {
1681 get_fs_pwd(current->fs, &nd->path);
1683 struct dentry *dentry;
1685 file = fget_light(dfd, &fput_needed);
1690 dentry = file->f_path.dentry;
1693 if (!S_ISDIR(dentry->d_inode->i_mode))
1696 retval = file_permission(file, MAY_EXEC);
1700 nd->path = file->f_path;
1701 path_get(&file->f_path);
1703 fput_light(file, fput_needed);
1705 nd->inode = nd->path.dentry->d_inode;
1709 fput_light(file, fput_needed);
1714 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1715 static int do_path_lookup(int dfd, const char *name,
1716 unsigned int flags, struct nameidata *nd)
1721 * Path walking is largely split up into 2 different synchronisation
1722 * schemes, rcu-walk and ref-walk (explained in
1723 * Documentation/filesystems/path-lookup.txt). These share much of the
1724 * path walk code, but some things particularly setup, cleanup, and
1725 * following mounts are sufficiently divergent that functions are
1726 * duplicated. Typically there is a function foo(), and its RCU
1727 * analogue, foo_rcu().
1729 * -ECHILD is the error number of choice (just to avoid clashes) that
1730 * is returned if some aspect of an rcu-walk fails. Such an error must
1731 * be handled by restarting a traditional ref-walk (which will always
1732 * be able to complete).
1734 retval = path_init_rcu(dfd, name, flags, nd);
1735 if (unlikely(retval))
1737 retval = path_walk_rcu(name, nd);
1738 path_finish_rcu(nd);
1740 path_put(&nd->root);
1741 nd->root.mnt = NULL;
1744 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1745 /* slower, locked walk */
1746 if (retval == -ESTALE)
1747 flags |= LOOKUP_REVAL;
1748 retval = path_init(dfd, name, flags, nd);
1749 if (unlikely(retval))
1751 retval = path_walk(name, nd);
1753 path_put(&nd->root);
1754 nd->root.mnt = NULL;
1758 if (likely(!retval)) {
1759 if (unlikely(!audit_dummy_context())) {
1760 if (nd->path.dentry && nd->inode)
1761 audit_inode(name, nd->path.dentry);
1768 int path_lookup(const char *name, unsigned int flags,
1769 struct nameidata *nd)
1771 return do_path_lookup(AT_FDCWD, name, flags, nd);
1774 int kern_path(const char *name, unsigned int flags, struct path *path)
1776 struct nameidata nd;
1777 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1784 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1785 * @dentry: pointer to dentry of the base directory
1786 * @mnt: pointer to vfs mount of the base directory
1787 * @name: pointer to file name
1788 * @flags: lookup flags
1789 * @nd: pointer to nameidata
1791 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1792 const char *name, unsigned int flags,
1793 struct nameidata *nd)
1797 /* same as do_path_lookup */
1798 nd->last_type = LAST_ROOT;
1802 nd->path.dentry = dentry;
1804 path_get(&nd->path);
1805 nd->root = nd->path;
1806 path_get(&nd->root);
1807 nd->inode = nd->path.dentry->d_inode;
1809 retval = path_walk(name, nd);
1810 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1812 audit_inode(name, nd->path.dentry);
1814 path_put(&nd->root);
1815 nd->root.mnt = NULL;
1820 static struct dentry *__lookup_hash(struct qstr *name,
1821 struct dentry *base, struct nameidata *nd)
1823 struct inode *inode = base->d_inode;
1824 struct dentry *dentry;
1827 err = exec_permission(inode, 0);
1829 return ERR_PTR(err);
1832 * See if the low-level filesystem might want
1833 * to use its own hash..
1835 if (base->d_flags & DCACHE_OP_HASH) {
1836 err = base->d_op->d_hash(base, inode, name);
1837 dentry = ERR_PTR(err);
1843 * Don't bother with __d_lookup: callers are for creat as
1844 * well as unlink, so a lot of the time it would cost
1847 dentry = d_lookup(base, name);
1849 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1850 dentry = do_revalidate(dentry, nd);
1853 dentry = d_alloc_and_lookup(base, name, nd);
1859 * Restricted form of lookup. Doesn't follow links, single-component only,
1860 * needs parent already locked. Doesn't follow mounts.
1863 static struct dentry *lookup_hash(struct nameidata *nd)
1865 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1868 static int __lookup_one_len(const char *name, struct qstr *this,
1869 struct dentry *base, int len)
1879 hash = init_name_hash();
1881 c = *(const unsigned char *)name++;
1882 if (c == '/' || c == '\0')
1884 hash = partial_name_hash(c, hash);
1886 this->hash = end_name_hash(hash);
1891 * lookup_one_len - filesystem helper to lookup single pathname component
1892 * @name: pathname component to lookup
1893 * @base: base directory to lookup from
1894 * @len: maximum length @len should be interpreted to
1896 * Note that this routine is purely a helper for filesystem usage and should
1897 * not be called by generic code. Also note that by using this function the
1898 * nameidata argument is passed to the filesystem methods and a filesystem
1899 * using this helper needs to be prepared for that.
1901 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1906 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1908 err = __lookup_one_len(name, &this, base, len);
1910 return ERR_PTR(err);
1912 return __lookup_hash(&this, base, NULL);
1915 int user_path_at(int dfd, const char __user *name, unsigned flags,
1918 struct nameidata nd;
1919 char *tmp = getname(name);
1920 int err = PTR_ERR(tmp);
1923 BUG_ON(flags & LOOKUP_PARENT);
1925 err = do_path_lookup(dfd, tmp, flags, &nd);
1933 static int user_path_parent(int dfd, const char __user *path,
1934 struct nameidata *nd, char **name)
1936 char *s = getname(path);
1942 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1952 * It's inline, so penalty for filesystems that don't use sticky bit is
1955 static inline int check_sticky(struct inode *dir, struct inode *inode)
1957 uid_t fsuid = current_fsuid();
1959 if (!(dir->i_mode & S_ISVTX))
1961 if (inode->i_uid == fsuid)
1963 if (dir->i_uid == fsuid)
1965 return !capable(CAP_FOWNER);
1969 * Check whether we can remove a link victim from directory dir, check
1970 * whether the type of victim is right.
1971 * 1. We can't do it if dir is read-only (done in permission())
1972 * 2. We should have write and exec permissions on dir
1973 * 3. We can't remove anything from append-only dir
1974 * 4. We can't do anything with immutable dir (done in permission())
1975 * 5. If the sticky bit on dir is set we should either
1976 * a. be owner of dir, or
1977 * b. be owner of victim, or
1978 * c. have CAP_FOWNER capability
1979 * 6. If the victim is append-only or immutable we can't do antyhing with
1980 * links pointing to it.
1981 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1982 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1983 * 9. We can't remove a root or mountpoint.
1984 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1985 * nfs_async_unlink().
1987 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1991 if (!victim->d_inode)
1994 BUG_ON(victim->d_parent->d_inode != dir);
1995 audit_inode_child(victim, dir);
1997 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2002 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2003 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2006 if (!S_ISDIR(victim->d_inode->i_mode))
2008 if (IS_ROOT(victim))
2010 } else if (S_ISDIR(victim->d_inode->i_mode))
2012 if (IS_DEADDIR(dir))
2014 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2019 /* Check whether we can create an object with dentry child in directory
2021 * 1. We can't do it if child already exists (open has special treatment for
2022 * this case, but since we are inlined it's OK)
2023 * 2. We can't do it if dir is read-only (done in permission())
2024 * 3. We should have write and exec permissions on dir
2025 * 4. We can't do it if dir is immutable (done in permission())
2027 static inline int may_create(struct inode *dir, struct dentry *child)
2031 if (IS_DEADDIR(dir))
2033 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2037 * p1 and p2 should be directories on the same fs.
2039 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2044 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2048 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2050 p = d_ancestor(p2, p1);
2052 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2053 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2057 p = d_ancestor(p1, p2);
2059 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2060 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2064 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2065 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2069 void unlock_rename(struct dentry *p1, struct dentry *p2)
2071 mutex_unlock(&p1->d_inode->i_mutex);
2073 mutex_unlock(&p2->d_inode->i_mutex);
2074 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2078 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
2079 struct nameidata *nd)
2081 int error = may_create(dir, dentry);
2086 if (!dir->i_op->create)
2087 return -EACCES; /* shouldn't it be ENOSYS? */
2090 error = security_inode_create(dir, dentry, mode);
2093 error = dir->i_op->create(dir, dentry, mode, nd);
2095 fsnotify_create(dir, dentry);
2099 int may_open(struct path *path, int acc_mode, int flag)
2101 struct dentry *dentry = path->dentry;
2102 struct inode *inode = dentry->d_inode;
2108 switch (inode->i_mode & S_IFMT) {
2112 if (acc_mode & MAY_WRITE)
2117 if (path->mnt->mnt_flags & MNT_NODEV)
2126 error = inode_permission(inode, acc_mode);
2131 * An append-only file must be opened in append mode for writing.
2133 if (IS_APPEND(inode)) {
2134 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2140 /* O_NOATIME can only be set by the owner or superuser */
2141 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2145 * Ensure there are no outstanding leases on the file.
2147 return break_lease(inode, flag);
2150 static int handle_truncate(struct file *filp)
2152 struct path *path = &filp->f_path;
2153 struct inode *inode = path->dentry->d_inode;
2154 int error = get_write_access(inode);
2158 * Refuse to truncate files with mandatory locks held on them.
2160 error = locks_verify_locked(inode);
2162 error = security_path_truncate(path);
2164 error = do_truncate(path->dentry, 0,
2165 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2168 put_write_access(inode);
2173 * Be careful about ever adding any more callers of this
2174 * function. Its flags must be in the namei format, not
2175 * what get passed to sys_open().
2177 static int __open_namei_create(struct nameidata *nd, struct path *path,
2178 int open_flag, int mode)
2181 struct dentry *dir = nd->path.dentry;
2183 if (!IS_POSIXACL(dir->d_inode))
2184 mode &= ~current_umask();
2185 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2188 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
2190 mutex_unlock(&dir->d_inode->i_mutex);
2191 dput(nd->path.dentry);
2192 nd->path.dentry = path->dentry;
2196 /* Don't check for write permission, don't truncate */
2197 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
2201 * Note that while the flag value (low two bits) for sys_open means:
2206 * it is changed into
2207 * 00 - no permissions needed
2208 * 01 - read-permission
2209 * 10 - write-permission
2211 * for the internal routines (ie open_namei()/follow_link() etc)
2212 * This is more logical, and also allows the 00 "no perm needed"
2213 * to be used for symlinks (where the permissions are checked
2217 static inline int open_to_namei_flags(int flag)
2219 if ((flag+1) & O_ACCMODE)
2224 static int open_will_truncate(int flag, struct inode *inode)
2227 * We'll never write to the fs underlying
2230 if (special_file(inode->i_mode))
2232 return (flag & O_TRUNC);
2235 static struct file *finish_open(struct nameidata *nd,
2236 int open_flag, int acc_mode)
2242 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
2243 if (will_truncate) {
2244 error = mnt_want_write(nd->path.mnt);
2248 error = may_open(&nd->path, acc_mode, open_flag);
2251 mnt_drop_write(nd->path.mnt);
2254 filp = nameidata_to_filp(nd);
2255 if (!IS_ERR(filp)) {
2256 error = ima_file_check(filp, acc_mode);
2259 filp = ERR_PTR(error);
2262 if (!IS_ERR(filp)) {
2263 if (will_truncate) {
2264 error = handle_truncate(filp);
2267 filp = ERR_PTR(error);
2272 * It is now safe to drop the mnt write
2273 * because the filp has had a write taken
2277 mnt_drop_write(nd->path.mnt);
2278 path_put(&nd->path);
2282 if (!IS_ERR(nd->intent.open.file))
2283 release_open_intent(nd);
2284 path_put(&nd->path);
2285 return ERR_PTR(error);
2289 * Handle O_CREAT case for do_filp_open
2291 static struct file *do_last(struct nameidata *nd, struct path *path,
2292 int open_flag, int acc_mode,
2293 int mode, const char *pathname)
2295 struct dentry *dir = nd->path.dentry;
2297 int error = -EISDIR;
2299 switch (nd->last_type) {
2302 dir = nd->path.dentry;
2304 if (need_reval_dot(dir)) {
2305 int status = d_revalidate(nd->path.dentry, nd);
2317 audit_inode(pathname, dir);
2321 /* trailing slashes? */
2322 if (nd->last.name[nd->last.len])
2325 mutex_lock(&dir->d_inode->i_mutex);
2327 path->dentry = lookup_hash(nd);
2328 path->mnt = nd->path.mnt;
2330 error = PTR_ERR(path->dentry);
2331 if (IS_ERR(path->dentry)) {
2332 mutex_unlock(&dir->d_inode->i_mutex);
2336 if (IS_ERR(nd->intent.open.file)) {
2337 error = PTR_ERR(nd->intent.open.file);
2338 goto exit_mutex_unlock;
2341 /* Negative dentry, just create the file */
2342 if (!path->dentry->d_inode) {
2344 * This write is needed to ensure that a
2345 * ro->rw transition does not occur between
2346 * the time when the file is created and when
2347 * a permanent write count is taken through
2348 * the 'struct file' in nameidata_to_filp().
2350 error = mnt_want_write(nd->path.mnt);
2352 goto exit_mutex_unlock;
2353 error = __open_namei_create(nd, path, open_flag, mode);
2355 mnt_drop_write(nd->path.mnt);
2358 filp = nameidata_to_filp(nd);
2359 mnt_drop_write(nd->path.mnt);
2360 path_put(&nd->path);
2361 if (!IS_ERR(filp)) {
2362 error = ima_file_check(filp, acc_mode);
2365 filp = ERR_PTR(error);
2372 * It already exists.
2374 mutex_unlock(&dir->d_inode->i_mutex);
2375 audit_inode(pathname, path->dentry);
2378 if (open_flag & O_EXCL)
2381 error = follow_managed(path, nd->flags);
2386 if (!path->dentry->d_inode)
2389 if (path->dentry->d_inode->i_op->follow_link)
2392 path_to_nameidata(path, nd);
2393 nd->inode = path->dentry->d_inode;
2395 if (S_ISDIR(nd->inode->i_mode))
2398 filp = finish_open(nd, open_flag, acc_mode);
2402 mutex_unlock(&dir->d_inode->i_mutex);
2404 path_put_conditional(path, nd);
2406 if (!IS_ERR(nd->intent.open.file))
2407 release_open_intent(nd);
2408 path_put(&nd->path);
2409 return ERR_PTR(error);
2413 * Note that the low bits of the passed in "open_flag"
2414 * are not the same as in the local variable "flag". See
2415 * open_to_namei_flags() for more details.
2417 struct file *do_filp_open(int dfd, const char *pathname,
2418 int open_flag, int mode, int acc_mode)
2421 struct nameidata nd;
2425 int flag = open_to_namei_flags(open_flag);
2428 if (!(open_flag & O_CREAT))
2431 /* Must never be set by userspace */
2432 open_flag &= ~FMODE_NONOTIFY;
2435 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2436 * check for O_DSYNC if the need any syncing at all we enforce it's
2437 * always set instead of having to deal with possibly weird behaviour
2438 * for malicious applications setting only __O_SYNC.
2440 if (open_flag & __O_SYNC)
2441 open_flag |= O_DSYNC;
2444 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
2446 /* O_TRUNC implies we need access checks for write permissions */
2447 if (open_flag & O_TRUNC)
2448 acc_mode |= MAY_WRITE;
2450 /* Allow the LSM permission hook to distinguish append
2451 access from general write access. */
2452 if (open_flag & O_APPEND)
2453 acc_mode |= MAY_APPEND;
2455 flags = LOOKUP_OPEN;
2456 if (open_flag & O_CREAT) {
2457 flags |= LOOKUP_CREATE;
2458 if (open_flag & O_EXCL)
2459 flags |= LOOKUP_EXCL;
2461 if (open_flag & O_DIRECTORY)
2462 flags |= LOOKUP_DIRECTORY;
2463 if (!(open_flag & O_NOFOLLOW))
2464 flags |= LOOKUP_FOLLOW;
2466 filp = get_empty_filp();
2468 return ERR_PTR(-ENFILE);
2470 filp->f_flags = open_flag;
2471 nd.intent.open.file = filp;
2472 nd.intent.open.flags = flag;
2473 nd.intent.open.create_mode = mode;
2475 if (open_flag & O_CREAT)
2478 /* !O_CREAT, simple open */
2479 error = do_path_lookup(dfd, pathname, flags, &nd);
2480 if (unlikely(error))
2483 if (!(nd.flags & LOOKUP_FOLLOW)) {
2484 if (nd.inode->i_op->follow_link)
2488 if (nd.flags & LOOKUP_DIRECTORY) {
2489 if (!nd.inode->i_op->lookup)
2492 audit_inode(pathname, nd.path.dentry);
2493 filp = finish_open(&nd, open_flag, acc_mode);
2497 /* OK, have to create the file. Find the parent. */
2498 error = path_init_rcu(dfd, pathname,
2499 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2502 error = path_walk_rcu(pathname, &nd);
2503 path_finish_rcu(&nd);
2504 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2505 /* slower, locked walk */
2506 if (error == -ESTALE) {
2508 flags |= LOOKUP_REVAL;
2510 error = path_init(dfd, pathname,
2511 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2515 error = path_walk_simple(pathname, &nd);
2517 if (unlikely(error))
2519 if (unlikely(!audit_dummy_context()))
2520 audit_inode(pathname, nd.path.dentry);
2523 * We have the parent and last component.
2526 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2527 while (unlikely(!filp)) { /* trailing symlink */
2528 struct path link = path;
2529 struct inode *linki = link.dentry->d_inode;
2532 if (!(nd.flags & LOOKUP_FOLLOW))
2537 * This is subtle. Instead of calling do_follow_link() we do
2538 * the thing by hands. The reason is that this way we have zero
2539 * link_count and path_walk() (called from ->follow_link)
2540 * honoring LOOKUP_PARENT. After that we have the parent and
2541 * last component, i.e. we are in the same situation as after
2542 * the first path_walk(). Well, almost - if the last component
2543 * is normal we get its copy stored in nd->last.name and we will
2544 * have to putname() it when we are done. Procfs-like symlinks
2545 * just set LAST_BIND.
2547 nd.flags |= LOOKUP_PARENT;
2548 error = security_inode_follow_link(link.dentry, &nd);
2551 error = __do_follow_link(&link, &nd, &cookie);
2552 if (unlikely(error)) {
2553 if (!IS_ERR(cookie) && linki->i_op->put_link)
2554 linki->i_op->put_link(link.dentry, &nd, cookie);
2555 /* nd.path had been dropped */
2559 nd.flags &= ~LOOKUP_PARENT;
2560 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2561 if (linki->i_op->put_link)
2562 linki->i_op->put_link(link.dentry, &nd, cookie);
2568 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
2573 path_put_conditional(&path, &nd);
2577 if (!IS_ERR(nd.intent.open.file))
2578 release_open_intent(&nd);
2579 filp = ERR_PTR(error);
2584 * filp_open - open file and return file pointer
2586 * @filename: path to open
2587 * @flags: open flags as per the open(2) second argument
2588 * @mode: mode for the new file if O_CREAT is set, else ignored
2590 * This is the helper to open a file from kernelspace if you really
2591 * have to. But in generally you should not do this, so please move
2592 * along, nothing to see here..
2594 struct file *filp_open(const char *filename, int flags, int mode)
2596 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
2598 EXPORT_SYMBOL(filp_open);
2601 * lookup_create - lookup a dentry, creating it if it doesn't exist
2602 * @nd: nameidata info
2603 * @is_dir: directory flag
2605 * Simple function to lookup and return a dentry and create it
2606 * if it doesn't exist. Is SMP-safe.
2608 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2610 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2612 struct dentry *dentry = ERR_PTR(-EEXIST);
2614 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2616 * Yucky last component or no last component at all?
2617 * (foo/., foo/.., /////)
2619 if (nd->last_type != LAST_NORM)
2621 nd->flags &= ~LOOKUP_PARENT;
2622 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2623 nd->intent.open.flags = O_EXCL;
2626 * Do the final lookup.
2628 dentry = lookup_hash(nd);
2632 if (dentry->d_inode)
2635 * Special case - lookup gave negative, but... we had foo/bar/
2636 * From the vfs_mknod() POV we just have a negative dentry -
2637 * all is fine. Let's be bastards - you had / on the end, you've
2638 * been asking for (non-existent) directory. -ENOENT for you.
2640 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2642 dentry = ERR_PTR(-ENOENT);
2647 dentry = ERR_PTR(-EEXIST);
2651 EXPORT_SYMBOL_GPL(lookup_create);
2653 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2655 int error = may_create(dir, dentry);
2660 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2663 if (!dir->i_op->mknod)
2666 error = devcgroup_inode_mknod(mode, dev);
2670 error = security_inode_mknod(dir, dentry, mode, dev);
2674 error = dir->i_op->mknod(dir, dentry, mode, dev);
2676 fsnotify_create(dir, dentry);
2680 static int may_mknod(mode_t mode)
2682 switch (mode & S_IFMT) {
2688 case 0: /* zero mode translates to S_IFREG */
2697 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2702 struct dentry *dentry;
2703 struct nameidata nd;
2708 error = user_path_parent(dfd, filename, &nd, &tmp);
2712 dentry = lookup_create(&nd, 0);
2713 if (IS_ERR(dentry)) {
2714 error = PTR_ERR(dentry);
2717 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2718 mode &= ~current_umask();
2719 error = may_mknod(mode);
2722 error = mnt_want_write(nd.path.mnt);
2725 error = security_path_mknod(&nd.path, dentry, mode, dev);
2727 goto out_drop_write;
2728 switch (mode & S_IFMT) {
2729 case 0: case S_IFREG:
2730 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2732 case S_IFCHR: case S_IFBLK:
2733 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2734 new_decode_dev(dev));
2736 case S_IFIFO: case S_IFSOCK:
2737 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2741 mnt_drop_write(nd.path.mnt);
2745 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2752 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2754 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2757 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2759 int error = may_create(dir, dentry);
2764 if (!dir->i_op->mkdir)
2767 mode &= (S_IRWXUGO|S_ISVTX);
2768 error = security_inode_mkdir(dir, dentry, mode);
2772 error = dir->i_op->mkdir(dir, dentry, mode);
2774 fsnotify_mkdir(dir, dentry);
2778 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2782 struct dentry *dentry;
2783 struct nameidata nd;
2785 error = user_path_parent(dfd, pathname, &nd, &tmp);
2789 dentry = lookup_create(&nd, 1);
2790 error = PTR_ERR(dentry);
2794 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2795 mode &= ~current_umask();
2796 error = mnt_want_write(nd.path.mnt);
2799 error = security_path_mkdir(&nd.path, dentry, mode);
2801 goto out_drop_write;
2802 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2804 mnt_drop_write(nd.path.mnt);
2808 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2815 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2817 return sys_mkdirat(AT_FDCWD, pathname, mode);
2821 * We try to drop the dentry early: we should have
2822 * a usage count of 2 if we're the only user of this
2823 * dentry, and if that is true (possibly after pruning
2824 * the dcache), then we drop the dentry now.
2826 * A low-level filesystem can, if it choses, legally
2829 * if (!d_unhashed(dentry))
2832 * if it cannot handle the case of removing a directory
2833 * that is still in use by something else..
2835 void dentry_unhash(struct dentry *dentry)
2838 shrink_dcache_parent(dentry);
2839 spin_lock(&dentry->d_lock);
2840 if (dentry->d_count == 2)
2842 spin_unlock(&dentry->d_lock);
2845 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2847 int error = may_delete(dir, dentry, 1);
2852 if (!dir->i_op->rmdir)
2855 mutex_lock(&dentry->d_inode->i_mutex);
2856 dentry_unhash(dentry);
2857 if (d_mountpoint(dentry))
2860 error = security_inode_rmdir(dir, dentry);
2862 error = dir->i_op->rmdir(dir, dentry);
2864 dentry->d_inode->i_flags |= S_DEAD;
2869 mutex_unlock(&dentry->d_inode->i_mutex);
2878 static long do_rmdir(int dfd, const char __user *pathname)
2882 struct dentry *dentry;
2883 struct nameidata nd;
2885 error = user_path_parent(dfd, pathname, &nd, &name);
2889 switch(nd.last_type) {
2901 nd.flags &= ~LOOKUP_PARENT;
2903 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2904 dentry = lookup_hash(&nd);
2905 error = PTR_ERR(dentry);
2908 error = mnt_want_write(nd.path.mnt);
2911 error = security_path_rmdir(&nd.path, dentry);
2914 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2916 mnt_drop_write(nd.path.mnt);
2920 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2927 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2929 return do_rmdir(AT_FDCWD, pathname);
2932 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2934 int error = may_delete(dir, dentry, 0);
2939 if (!dir->i_op->unlink)
2942 mutex_lock(&dentry->d_inode->i_mutex);
2943 if (d_mountpoint(dentry))
2946 error = security_inode_unlink(dir, dentry);
2948 error = dir->i_op->unlink(dir, dentry);
2953 mutex_unlock(&dentry->d_inode->i_mutex);
2955 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2956 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2957 fsnotify_link_count(dentry->d_inode);
2965 * Make sure that the actual truncation of the file will occur outside its
2966 * directory's i_mutex. Truncate can take a long time if there is a lot of
2967 * writeout happening, and we don't want to prevent access to the directory
2968 * while waiting on the I/O.
2970 static long do_unlinkat(int dfd, const char __user *pathname)
2974 struct dentry *dentry;
2975 struct nameidata nd;
2976 struct inode *inode = NULL;
2978 error = user_path_parent(dfd, pathname, &nd, &name);
2983 if (nd.last_type != LAST_NORM)
2986 nd.flags &= ~LOOKUP_PARENT;
2988 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2989 dentry = lookup_hash(&nd);
2990 error = PTR_ERR(dentry);
2991 if (!IS_ERR(dentry)) {
2992 /* Why not before? Because we want correct error value */
2993 if (nd.last.name[nd.last.len])
2995 inode = dentry->d_inode;
2998 error = mnt_want_write(nd.path.mnt);
3001 error = security_path_unlink(&nd.path, dentry);
3004 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3006 mnt_drop_write(nd.path.mnt);
3010 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3012 iput(inode); /* truncate the inode here */
3019 error = !dentry->d_inode ? -ENOENT :
3020 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3024 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3026 if ((flag & ~AT_REMOVEDIR) != 0)
3029 if (flag & AT_REMOVEDIR)
3030 return do_rmdir(dfd, pathname);
3032 return do_unlinkat(dfd, pathname);
3035 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3037 return do_unlinkat(AT_FDCWD, pathname);
3040 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3042 int error = may_create(dir, dentry);
3047 if (!dir->i_op->symlink)
3050 error = security_inode_symlink(dir, dentry, oldname);
3054 error = dir->i_op->symlink(dir, dentry, oldname);
3056 fsnotify_create(dir, dentry);
3060 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3061 int, newdfd, const char __user *, newname)
3066 struct dentry *dentry;
3067 struct nameidata nd;
3069 from = getname(oldname);
3071 return PTR_ERR(from);
3073 error = user_path_parent(newdfd, newname, &nd, &to);
3077 dentry = lookup_create(&nd, 0);
3078 error = PTR_ERR(dentry);
3082 error = mnt_want_write(nd.path.mnt);
3085 error = security_path_symlink(&nd.path, dentry, from);
3087 goto out_drop_write;
3088 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
3090 mnt_drop_write(nd.path.mnt);
3094 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3102 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3104 return sys_symlinkat(oldname, AT_FDCWD, newname);
3107 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3109 struct inode *inode = old_dentry->d_inode;
3115 error = may_create(dir, new_dentry);
3119 if (dir->i_sb != inode->i_sb)
3123 * A link to an append-only or immutable file cannot be created.
3125 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3127 if (!dir->i_op->link)
3129 if (S_ISDIR(inode->i_mode))
3132 error = security_inode_link(old_dentry, dir, new_dentry);
3136 mutex_lock(&inode->i_mutex);
3137 error = dir->i_op->link(old_dentry, dir, new_dentry);
3138 mutex_unlock(&inode->i_mutex);
3140 fsnotify_link(dir, inode, new_dentry);
3145 * Hardlinks are often used in delicate situations. We avoid
3146 * security-related surprises by not following symlinks on the
3149 * We don't follow them on the oldname either to be compatible
3150 * with linux 2.0, and to avoid hard-linking to directories
3151 * and other special files. --ADM
3153 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3154 int, newdfd, const char __user *, newname, int, flags)
3156 struct dentry *new_dentry;
3157 struct nameidata nd;
3158 struct path old_path;
3162 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
3165 error = user_path_at(olddfd, oldname,
3166 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
3171 error = user_path_parent(newdfd, newname, &nd, &to);
3175 if (old_path.mnt != nd.path.mnt)
3177 new_dentry = lookup_create(&nd, 0);
3178 error = PTR_ERR(new_dentry);
3179 if (IS_ERR(new_dentry))
3181 error = mnt_want_write(nd.path.mnt);
3184 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3186 goto out_drop_write;
3187 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
3189 mnt_drop_write(nd.path.mnt);
3193 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3198 path_put(&old_path);
3203 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3205 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3209 * The worst of all namespace operations - renaming directory. "Perverted"
3210 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3212 * a) we can get into loop creation. Check is done in is_subdir().
3213 * b) race potential - two innocent renames can create a loop together.
3214 * That's where 4.4 screws up. Current fix: serialization on
3215 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3217 * c) we have to lock _three_ objects - parents and victim (if it exists).
3218 * And that - after we got ->i_mutex on parents (until then we don't know
3219 * whether the target exists). Solution: try to be smart with locking
3220 * order for inodes. We rely on the fact that tree topology may change
3221 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3222 * move will be locked. Thus we can rank directories by the tree
3223 * (ancestors first) and rank all non-directories after them.
3224 * That works since everybody except rename does "lock parent, lookup,
3225 * lock child" and rename is under ->s_vfs_rename_mutex.
3226 * HOWEVER, it relies on the assumption that any object with ->lookup()
3227 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3228 * we'd better make sure that there's no link(2) for them.
3229 * d) some filesystems don't support opened-but-unlinked directories,
3230 * either because of layout or because they are not ready to deal with
3231 * all cases correctly. The latter will be fixed (taking this sort of
3232 * stuff into VFS), but the former is not going away. Solution: the same
3233 * trick as in rmdir().
3234 * e) conversion from fhandle to dentry may come in the wrong moment - when
3235 * we are removing the target. Solution: we will have to grab ->i_mutex
3236 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3237 * ->i_mutex on parents, which works but leads to some truly excessive
3240 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3241 struct inode *new_dir, struct dentry *new_dentry)
3244 struct inode *target;
3247 * If we are going to change the parent - check write permissions,
3248 * we'll need to flip '..'.
3250 if (new_dir != old_dir) {
3251 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3256 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3260 target = new_dentry->d_inode;
3262 mutex_lock(&target->i_mutex);
3263 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3267 dentry_unhash(new_dentry);
3268 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3272 target->i_flags |= S_DEAD;
3273 dont_mount(new_dentry);
3275 mutex_unlock(&target->i_mutex);
3276 if (d_unhashed(new_dentry))
3277 d_rehash(new_dentry);
3281 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3282 d_move(old_dentry,new_dentry);
3286 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3287 struct inode *new_dir, struct dentry *new_dentry)
3289 struct inode *target;
3292 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3297 target = new_dentry->d_inode;
3299 mutex_lock(&target->i_mutex);
3300 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3303 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3306 dont_mount(new_dentry);
3307 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3308 d_move(old_dentry, new_dentry);
3311 mutex_unlock(&target->i_mutex);
3316 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3317 struct inode *new_dir, struct dentry *new_dentry)
3320 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3321 const unsigned char *old_name;
3323 if (old_dentry->d_inode == new_dentry->d_inode)
3326 error = may_delete(old_dir, old_dentry, is_dir);
3330 if (!new_dentry->d_inode)
3331 error = may_create(new_dir, new_dentry);
3333 error = may_delete(new_dir, new_dentry, is_dir);
3337 if (!old_dir->i_op->rename)
3340 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3343 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3345 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3347 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3348 new_dentry->d_inode, old_dentry);
3349 fsnotify_oldname_free(old_name);
3354 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3355 int, newdfd, const char __user *, newname)
3357 struct dentry *old_dir, *new_dir;
3358 struct dentry *old_dentry, *new_dentry;
3359 struct dentry *trap;
3360 struct nameidata oldnd, newnd;
3365 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3369 error = user_path_parent(newdfd, newname, &newnd, &to);
3374 if (oldnd.path.mnt != newnd.path.mnt)
3377 old_dir = oldnd.path.dentry;
3379 if (oldnd.last_type != LAST_NORM)
3382 new_dir = newnd.path.dentry;
3383 if (newnd.last_type != LAST_NORM)
3386 oldnd.flags &= ~LOOKUP_PARENT;
3387 newnd.flags &= ~LOOKUP_PARENT;
3388 newnd.flags |= LOOKUP_RENAME_TARGET;
3390 trap = lock_rename(new_dir, old_dir);
3392 old_dentry = lookup_hash(&oldnd);
3393 error = PTR_ERR(old_dentry);
3394 if (IS_ERR(old_dentry))
3396 /* source must exist */
3398 if (!old_dentry->d_inode)
3400 /* unless the source is a directory trailing slashes give -ENOTDIR */
3401 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3403 if (oldnd.last.name[oldnd.last.len])
3405 if (newnd.last.name[newnd.last.len])
3408 /* source should not be ancestor of target */
3410 if (old_dentry == trap)
3412 new_dentry = lookup_hash(&newnd);
3413 error = PTR_ERR(new_dentry);
3414 if (IS_ERR(new_dentry))
3416 /* target should not be an ancestor of source */
3418 if (new_dentry == trap)
3421 error = mnt_want_write(oldnd.path.mnt);
3424 error = security_path_rename(&oldnd.path, old_dentry,
3425 &newnd.path, new_dentry);
3428 error = vfs_rename(old_dir->d_inode, old_dentry,
3429 new_dir->d_inode, new_dentry);
3431 mnt_drop_write(oldnd.path.mnt);
3437 unlock_rename(new_dir, old_dir);
3439 path_put(&newnd.path);
3442 path_put(&oldnd.path);
3448 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3450 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3453 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3457 len = PTR_ERR(link);
3462 if (len > (unsigned) buflen)
3464 if (copy_to_user(buffer, link, len))
3471 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3472 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3473 * using) it for any given inode is up to filesystem.
3475 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3477 struct nameidata nd;
3482 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3484 return PTR_ERR(cookie);
3486 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3487 if (dentry->d_inode->i_op->put_link)
3488 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3492 int vfs_follow_link(struct nameidata *nd, const char *link)
3494 return __vfs_follow_link(nd, link);
3497 /* get the link contents into pagecache */
3498 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3502 struct address_space *mapping = dentry->d_inode->i_mapping;
3503 page = read_mapping_page(mapping, 0, NULL);
3508 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3512 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3514 struct page *page = NULL;
3515 char *s = page_getlink(dentry, &page);
3516 int res = vfs_readlink(dentry,buffer,buflen,s);
3519 page_cache_release(page);
3524 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3526 struct page *page = NULL;
3527 nd_set_link(nd, page_getlink(dentry, &page));
3531 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3533 struct page *page = cookie;
3537 page_cache_release(page);
3542 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3544 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3546 struct address_space *mapping = inode->i_mapping;
3551 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3553 flags |= AOP_FLAG_NOFS;
3556 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3557 flags, &page, &fsdata);
3561 kaddr = kmap_atomic(page, KM_USER0);
3562 memcpy(kaddr, symname, len-1);
3563 kunmap_atomic(kaddr, KM_USER0);
3565 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3572 mark_inode_dirty(inode);
3578 int page_symlink(struct inode *inode, const char *symname, int len)
3580 return __page_symlink(inode, symname, len,
3581 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3584 const struct inode_operations page_symlink_inode_operations = {
3585 .readlink = generic_readlink,
3586 .follow_link = page_follow_link_light,
3587 .put_link = page_put_link,
3590 EXPORT_SYMBOL(user_path_at);
3591 EXPORT_SYMBOL(follow_down_one);
3592 EXPORT_SYMBOL(follow_down);
3593 EXPORT_SYMBOL(follow_up);
3594 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3595 EXPORT_SYMBOL(getname);
3596 EXPORT_SYMBOL(lock_rename);
3597 EXPORT_SYMBOL(lookup_one_len);
3598 EXPORT_SYMBOL(page_follow_link_light);
3599 EXPORT_SYMBOL(page_put_link);
3600 EXPORT_SYMBOL(page_readlink);
3601 EXPORT_SYMBOL(__page_symlink);
3602 EXPORT_SYMBOL(page_symlink);
3603 EXPORT_SYMBOL(page_symlink_inode_operations);
3604 EXPORT_SYMBOL(path_lookup);
3605 EXPORT_SYMBOL(kern_path);
3606 EXPORT_SYMBOL(vfs_path_lookup);
3607 EXPORT_SYMBOL(inode_permission);
3608 EXPORT_SYMBOL(file_permission);
3609 EXPORT_SYMBOL(unlock_rename);
3610 EXPORT_SYMBOL(vfs_create);
3611 EXPORT_SYMBOL(vfs_follow_link);
3612 EXPORT_SYMBOL(vfs_link);
3613 EXPORT_SYMBOL(vfs_mkdir);
3614 EXPORT_SYMBOL(vfs_mknod);
3615 EXPORT_SYMBOL(generic_permission);
3616 EXPORT_SYMBOL(vfs_readlink);
3617 EXPORT_SYMBOL(vfs_rename);
3618 EXPORT_SYMBOL(vfs_rmdir);
3619 EXPORT_SYMBOL(vfs_symlink);
3620 EXPORT_SYMBOL(vfs_unlink);
3621 EXPORT_SYMBOL(dentry_unhash);
3622 EXPORT_SYMBOL(generic_readlink);