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_put - put a reference to a path
372 * @path: path to put the reference to
374 * Given a path decrement the reference count to the dentry and the vfsmount.
376 void path_put(struct path *path)
381 EXPORT_SYMBOL(path_put);
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
386 * Returns: 0 on success, -ECHILD on failure
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
400 static int nameidata_drop_rcu(struct nameidata *nd)
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
405 BUG_ON(!(nd->flags & LOOKUP_RCU));
407 spin_lock(&fs->lock);
408 if (nd->root.mnt != fs->root.mnt ||
409 nd->root.dentry != fs->root.dentry)
412 spin_lock(&dentry->d_lock);
413 if (!__d_rcu_to_refcount(dentry, nd->seq))
415 BUG_ON(nd->inode != dentry->d_inode);
416 spin_unlock(&dentry->d_lock);
419 spin_unlock(&fs->lock);
421 mntget(nd->path.mnt);
424 br_read_unlock(vfsmount_lock);
425 nd->flags &= ~LOOKUP_RCU;
428 spin_unlock(&dentry->d_lock);
431 spin_unlock(&fs->lock);
435 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
438 if (nd->flags & LOOKUP_RCU)
439 return nameidata_drop_rcu(nd);
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
447 * Returns: 0 on success, -ECHILD on failure
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
453 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
460 spin_lock(&fs->lock);
461 if (nd->root.mnt != fs->root.mnt ||
462 nd->root.dentry != fs->root.dentry)
465 spin_lock(&parent->d_lock);
466 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
467 if (!__d_rcu_to_refcount(dentry, nd->seq))
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
474 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
475 BUG_ON(!parent->d_count);
477 spin_unlock(&dentry->d_lock);
478 spin_unlock(&parent->d_lock);
481 spin_unlock(&fs->lock);
483 mntget(nd->path.mnt);
486 br_read_unlock(vfsmount_lock);
487 nd->flags &= ~LOOKUP_RCU;
490 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
494 spin_unlock(&fs->lock);
498 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
501 if (nd->flags & LOOKUP_RCU)
502 return nameidata_dentry_drop_rcu(nd, dentry);
507 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
508 * @nd: nameidata pathwalk data to drop
509 * Returns: 0 on success, -ECHILD on failure
511 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
512 * nd->path should be the final element of the lookup, so nd->root is discarded.
513 * Must be called from rcu-walk context.
515 static int nameidata_drop_rcu_last(struct nameidata *nd)
517 struct dentry *dentry = nd->path.dentry;
519 BUG_ON(!(nd->flags & LOOKUP_RCU));
520 nd->flags &= ~LOOKUP_RCU;
522 spin_lock(&dentry->d_lock);
523 if (!__d_rcu_to_refcount(dentry, nd->seq))
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
531 br_read_unlock(vfsmount_lock);
536 spin_unlock(&dentry->d_lock);
538 br_read_unlock(vfsmount_lock);
542 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
543 static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
545 if (likely(nd->flags & LOOKUP_RCU))
546 return nameidata_drop_rcu_last(nd);
551 * release_open_intent - free up open intent resources
552 * @nd: pointer to nameidata
554 void release_open_intent(struct nameidata *nd)
556 struct file *file = nd->intent.open.file;
558 if (file && !IS_ERR(file)) {
559 if (file->f_path.dentry == NULL)
566 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
568 return dentry->d_op->d_revalidate(dentry, nd);
571 static struct dentry *
572 do_revalidate(struct dentry *dentry, struct nameidata *nd)
574 int status = d_revalidate(dentry, nd);
575 if (unlikely(status <= 0)) {
577 * The dentry failed validation.
578 * If d_revalidate returned 0 attempt to invalidate
579 * the dentry otherwise d_revalidate is asking us
580 * to return a fail status.
584 dentry = ERR_PTR(status);
585 } else if (!d_invalidate(dentry)) {
593 static inline struct dentry *
594 do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
596 int status = d_revalidate(dentry, nd);
597 if (likely(status > 0))
599 if (status == -ECHILD) {
600 if (nameidata_dentry_drop_rcu(nd, dentry))
601 return ERR_PTR(-ECHILD);
602 return do_revalidate(dentry, nd);
605 return ERR_PTR(status);
606 /* Don't d_invalidate in rcu-walk mode */
607 if (nameidata_dentry_drop_rcu(nd, dentry))
608 return ERR_PTR(-ECHILD);
609 if (!d_invalidate(dentry)) {
616 static inline int need_reval_dot(struct dentry *dentry)
618 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
621 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
628 * force_reval_path - force revalidation of a dentry
630 * In some situations the path walking code will trust dentries without
631 * revalidating them. This causes problems for filesystems that depend on
632 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
633 * (which indicates that it's possible for the dentry to go stale), force
634 * a d_revalidate call before proceeding.
636 * Returns 0 if the revalidation was successful. If the revalidation fails,
637 * either return the error returned by d_revalidate or -ESTALE if the
638 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
639 * invalidate the dentry. It's up to the caller to handle putting references
640 * to the path if necessary.
643 force_reval_path(struct path *path, struct nameidata *nd)
646 struct dentry *dentry = path->dentry;
649 * only check on filesystems where it's possible for the dentry to
652 if (!need_reval_dot(dentry))
655 status = d_revalidate(dentry, nd);
660 d_invalidate(dentry);
667 * Short-cut version of permission(), for calling on directories
668 * during pathname resolution. Combines parts of permission()
669 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
671 * If appropriate, check DAC only. If not appropriate, or
672 * short-cut DAC fails, then call ->permission() to do more
673 * complete permission check.
675 static inline int exec_permission(struct inode *inode, unsigned int flags)
679 if (inode->i_op->permission) {
680 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
682 ret = acl_permission_check(inode, MAY_EXEC, flags,
683 inode->i_op->check_acl);
690 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
695 return security_inode_exec_permission(inode, flags);
698 static __always_inline void set_root(struct nameidata *nd)
701 get_fs_root(current->fs, &nd->root);
704 static int link_path_walk(const char *, struct nameidata *);
706 static __always_inline void set_root_rcu(struct nameidata *nd)
709 struct fs_struct *fs = current->fs;
713 seq = read_seqcount_begin(&fs->seq);
715 } while (read_seqcount_retry(&fs->seq, seq));
719 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
732 nd->inode = nd->path.dentry->d_inode;
734 ret = link_path_walk(link, nd);
738 return PTR_ERR(link);
741 static void path_put_conditional(struct path *path, struct nameidata *nd)
744 if (path->mnt != nd->path.mnt)
748 static inline void path_to_nameidata(const struct path *path,
749 struct nameidata *nd)
751 if (!(nd->flags & LOOKUP_RCU)) {
752 dput(nd->path.dentry);
753 if (nd->path.mnt != path->mnt)
754 mntput(nd->path.mnt);
756 nd->path.mnt = path->mnt;
757 nd->path.dentry = path->dentry;
760 static __always_inline int
761 __do_follow_link(const struct path *link, struct nameidata *nd, void **p)
764 struct dentry *dentry = link->dentry;
766 BUG_ON(nd->flags & LOOKUP_RCU);
768 touch_atime(link->mnt, dentry);
769 nd_set_link(nd, NULL);
771 if (link->mnt == nd->path.mnt)
774 nd->last_type = LAST_BIND;
775 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
778 char *s = nd_get_link(nd);
781 error = __vfs_follow_link(nd, s);
782 else if (nd->last_type == LAST_BIND) {
783 error = force_reval_path(&nd->path, nd);
792 * This limits recursive symlink follows to 8, while
793 * limiting consecutive symlinks to 40.
795 * Without that kind of total limit, nasty chains of consecutive
796 * symlinks can cause almost arbitrarily long lookups.
798 static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
803 /* We drop rcu-walk here */
804 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
806 BUG_ON(inode != path->dentry->d_inode);
808 if (current->link_count >= MAX_NESTED_LINKS)
810 if (current->total_link_count >= 40)
812 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
814 err = security_inode_follow_link(path->dentry, nd);
817 current->link_count++;
818 current->total_link_count++;
820 err = __do_follow_link(path, nd, &cookie);
821 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
822 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
824 current->link_count--;
828 path_put_conditional(path, nd);
833 static int follow_up_rcu(struct path *path)
835 struct vfsmount *parent;
836 struct dentry *mountpoint;
838 parent = path->mnt->mnt_parent;
839 if (parent == path->mnt)
841 mountpoint = path->mnt->mnt_mountpoint;
842 path->dentry = mountpoint;
847 int follow_up(struct path *path)
849 struct vfsmount *parent;
850 struct dentry *mountpoint;
852 br_read_lock(vfsmount_lock);
853 parent = path->mnt->mnt_parent;
854 if (parent == path->mnt) {
855 br_read_unlock(vfsmount_lock);
859 mountpoint = dget(path->mnt->mnt_mountpoint);
860 br_read_unlock(vfsmount_lock);
862 path->dentry = mountpoint;
869 * Perform an automount
870 * - return -EISDIR to tell follow_managed() to stop and return the path we
873 static int follow_automount(struct path *path, unsigned flags,
876 struct vfsmount *mnt;
879 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
882 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
883 * and this is the terminal part of the path.
885 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
886 return -EISDIR; /* we actually want to stop here */
888 /* We want to mount if someone is trying to open/create a file of any
889 * type under the mountpoint, wants to traverse through the mountpoint
890 * or wants to open the mounted directory.
892 * We don't want to mount if someone's just doing a stat and they've
893 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
894 * appended a '/' to the name.
896 if (!(flags & LOOKUP_FOLLOW) &&
897 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
898 LOOKUP_OPEN | LOOKUP_CREATE)))
901 current->total_link_count++;
902 if (current->total_link_count >= 40)
905 mnt = path->dentry->d_op->d_automount(path);
908 * The filesystem is allowed to return -EISDIR here to indicate
909 * it doesn't want to automount. For instance, autofs would do
910 * this so that its userspace daemon can mount on this dentry.
912 * However, we can only permit this if it's a terminal point in
913 * the path being looked up; if it wasn't then the remainder of
914 * the path is inaccessible and we should say so.
916 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
921 if (!mnt) /* mount collision */
924 err = finish_automount(mnt, path);
928 /* Someone else made a mount here whilst we were busy */
935 path->dentry = dget(mnt->mnt_root);
945 * Handle a dentry that is managed in some way.
946 * - Flagged for transit management (autofs)
947 * - Flagged as mountpoint
948 * - Flagged as automount point
950 * This may only be called in refwalk mode.
952 * Serialization is taken care of in namespace.c
954 static int follow_managed(struct path *path, unsigned flags)
957 bool need_mntput = false;
960 /* Given that we're not holding a lock here, we retain the value in a
961 * local variable for each dentry as we look at it so that we don't see
962 * the components of that value change under us */
963 while (managed = ACCESS_ONCE(path->dentry->d_flags),
964 managed &= DCACHE_MANAGED_DENTRY,
965 unlikely(managed != 0)) {
966 /* Allow the filesystem to manage the transit without i_mutex
968 if (managed & DCACHE_MANAGE_TRANSIT) {
969 BUG_ON(!path->dentry->d_op);
970 BUG_ON(!path->dentry->d_op->d_manage);
971 ret = path->dentry->d_op->d_manage(path->dentry,
974 return ret == -EISDIR ? 0 : ret;
977 /* Transit to a mounted filesystem. */
978 if (managed & DCACHE_MOUNTED) {
979 struct vfsmount *mounted = lookup_mnt(path);
985 path->dentry = dget(mounted->mnt_root);
990 /* Something is mounted on this dentry in another
991 * namespace and/or whatever was mounted there in this
992 * namespace got unmounted before we managed to get the
996 /* Handle an automount point */
997 if (managed & DCACHE_NEED_AUTOMOUNT) {
998 ret = follow_automount(path, flags, &need_mntput);
1000 return ret == -EISDIR ? 0 : ret;
1004 /* We didn't change the current path point */
1010 int follow_down_one(struct path *path)
1012 struct vfsmount *mounted;
1014 mounted = lookup_mnt(path);
1018 path->mnt = mounted;
1019 path->dentry = dget(mounted->mnt_root);
1026 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1027 * meet a managed dentry and we're not walking to "..". True is returned to
1028 * continue, false to abort.
1030 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1031 struct inode **inode, bool reverse_transit)
1033 while (d_mountpoint(path->dentry)) {
1034 struct vfsmount *mounted;
1035 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1037 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1039 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1042 path->mnt = mounted;
1043 path->dentry = mounted->mnt_root;
1044 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1045 *inode = path->dentry->d_inode;
1048 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1049 return reverse_transit;
1053 static int follow_dotdot_rcu(struct nameidata *nd)
1055 struct inode *inode = nd->inode;
1060 if (nd->path.dentry == nd->root.dentry &&
1061 nd->path.mnt == nd->root.mnt) {
1064 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1065 struct dentry *old = nd->path.dentry;
1066 struct dentry *parent = old->d_parent;
1069 seq = read_seqcount_begin(&parent->d_seq);
1070 if (read_seqcount_retry(&old->d_seq, nd->seq))
1072 inode = parent->d_inode;
1073 nd->path.dentry = parent;
1077 if (!follow_up_rcu(&nd->path))
1079 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1080 inode = nd->path.dentry->d_inode;
1082 __follow_mount_rcu(nd, &nd->path, &inode, true);
1089 * Follow down to the covering mount currently visible to userspace. At each
1090 * point, the filesystem owning that dentry may be queried as to whether the
1091 * caller is permitted to proceed or not.
1093 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1096 int follow_down(struct path *path, bool mounting_here)
1101 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1102 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1103 /* Allow the filesystem to manage the transit without i_mutex
1106 * We indicate to the filesystem if someone is trying to mount
1107 * something here. This gives autofs the chance to deny anyone
1108 * other than its daemon the right to mount on its
1111 * The filesystem may sleep at this point.
1113 if (managed & DCACHE_MANAGE_TRANSIT) {
1114 BUG_ON(!path->dentry->d_op);
1115 BUG_ON(!path->dentry->d_op->d_manage);
1116 ret = path->dentry->d_op->d_manage(
1117 path->dentry, mounting_here, false);
1119 return ret == -EISDIR ? 0 : ret;
1122 /* Transit to a mounted filesystem. */
1123 if (managed & DCACHE_MOUNTED) {
1124 struct vfsmount *mounted = lookup_mnt(path);
1129 path->mnt = mounted;
1130 path->dentry = dget(mounted->mnt_root);
1134 /* Don't handle automount points here */
1141 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1143 static void follow_mount(struct path *path)
1145 while (d_mountpoint(path->dentry)) {
1146 struct vfsmount *mounted = lookup_mnt(path);
1151 path->mnt = mounted;
1152 path->dentry = dget(mounted->mnt_root);
1156 static void follow_dotdot(struct nameidata *nd)
1161 struct dentry *old = nd->path.dentry;
1163 if (nd->path.dentry == nd->root.dentry &&
1164 nd->path.mnt == nd->root.mnt) {
1167 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1168 /* rare case of legitimate dget_parent()... */
1169 nd->path.dentry = dget_parent(nd->path.dentry);
1173 if (!follow_up(&nd->path))
1176 follow_mount(&nd->path);
1177 nd->inode = nd->path.dentry->d_inode;
1181 * Allocate a dentry with name and parent, and perform a parent
1182 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1183 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1184 * have verified that no child exists while under i_mutex.
1186 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1187 struct qstr *name, struct nameidata *nd)
1189 struct inode *inode = parent->d_inode;
1190 struct dentry *dentry;
1193 /* Don't create child dentry for a dead directory. */
1194 if (unlikely(IS_DEADDIR(inode)))
1195 return ERR_PTR(-ENOENT);
1197 dentry = d_alloc(parent, name);
1198 if (unlikely(!dentry))
1199 return ERR_PTR(-ENOMEM);
1201 old = inode->i_op->lookup(inode, dentry, nd);
1202 if (unlikely(old)) {
1210 * It's more convoluted than I'd like it to be, but... it's still fairly
1211 * small and for now I'd prefer to have fast path as straight as possible.
1212 * It _is_ time-critical.
1214 static int do_lookup(struct nameidata *nd, struct qstr *name,
1215 struct path *path, struct inode **inode)
1217 struct vfsmount *mnt = nd->path.mnt;
1218 struct dentry *dentry, *parent = nd->path.dentry;
1223 * See if the low-level filesystem might want
1224 * to use its own hash..
1226 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1227 err = parent->d_op->d_hash(parent, nd->inode, name);
1233 * Rename seqlock is not required here because in the off chance
1234 * of a false negative due to a concurrent rename, we're going to
1235 * do the non-racy lookup, below.
1237 if (nd->flags & LOOKUP_RCU) {
1241 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1243 if (nameidata_drop_rcu(nd))
1247 /* Memory barrier in read_seqcount_begin of child is enough */
1248 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1252 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1253 dentry = do_revalidate_rcu(dentry, nd);
1258 if (!(nd->flags & LOOKUP_RCU))
1262 path->dentry = dentry;
1263 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1265 if (nameidata_drop_rcu(nd))
1269 dentry = __d_lookup(parent, name);
1273 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1274 dentry = do_revalidate(dentry, nd);
1282 path->dentry = dentry;
1283 err = follow_managed(path, nd->flags);
1284 if (unlikely(err < 0)) {
1285 path_put_conditional(path, nd);
1288 *inode = path->dentry->d_inode;
1292 dir = parent->d_inode;
1293 BUG_ON(nd->inode != dir);
1295 mutex_lock(&dir->i_mutex);
1297 * First re-do the cached lookup just in case it was created
1298 * while we waited for the directory semaphore, or the first
1299 * lookup failed due to an unrelated rename.
1301 * This could use version numbering or similar to avoid unnecessary
1302 * cache lookups, but then we'd have to do the first lookup in the
1303 * non-racy way. However in the common case here, everything should
1304 * be hot in cache, so would it be a big win?
1306 dentry = d_lookup(parent, name);
1307 if (likely(!dentry)) {
1308 dentry = d_alloc_and_lookup(parent, name, nd);
1309 mutex_unlock(&dir->i_mutex);
1315 * Uhhuh! Nasty case: the cache was re-populated while
1316 * we waited on the semaphore. Need to revalidate.
1318 mutex_unlock(&dir->i_mutex);
1322 return PTR_ERR(dentry);
1327 * This is the basic name resolution function, turning a pathname into
1328 * the final dentry. We expect 'base' to be positive and a directory.
1330 * Returns 0 and nd will have valid dentry and mnt on success.
1331 * Returns error and drops reference to input namei data on failure.
1333 static int link_path_walk(const char *name, struct nameidata *nd)
1337 unsigned int lookup_flags = nd->flags;
1345 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1347 /* At this point we know we have a real path component. */
1349 struct inode *inode;
1354 nd->flags |= LOOKUP_CONTINUE;
1355 if (nd->flags & LOOKUP_RCU) {
1356 err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1357 if (err == -ECHILD) {
1358 if (nameidata_drop_rcu(nd))
1364 err = exec_permission(nd->inode, 0);
1370 c = *(const unsigned char *)name;
1372 hash = init_name_hash();
1375 hash = partial_name_hash(c, hash);
1376 c = *(const unsigned char *)name;
1377 } while (c && (c != '/'));
1378 this.len = name - (const char *) this.name;
1379 this.hash = end_name_hash(hash);
1381 /* remove trailing slashes? */
1383 goto last_component;
1384 while (*++name == '/');
1386 goto last_with_slashes;
1389 * "." and ".." are special - ".." especially so because it has
1390 * to be able to know about the current root directory and
1391 * parent relationships.
1393 if (this.name[0] == '.') switch (this.len) {
1397 if (this.name[1] != '.')
1399 if (nd->flags & LOOKUP_RCU) {
1400 if (follow_dotdot_rcu(nd))
1408 /* This does the actual lookups.. */
1409 err = do_lookup(nd, &this, &next, &inode);
1416 if (inode->i_op->follow_link) {
1417 err = do_follow_link(inode, &next, nd);
1420 nd->inode = nd->path.dentry->d_inode;
1425 path_to_nameidata(&next, nd);
1429 if (!nd->inode->i_op->lookup)
1432 /* here ends the main loop */
1435 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1437 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1438 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1439 if (lookup_flags & LOOKUP_PARENT)
1441 if (this.name[0] == '.') switch (this.len) {
1445 if (this.name[1] != '.')
1447 if (nd->flags & LOOKUP_RCU) {
1448 if (follow_dotdot_rcu(nd))
1456 err = do_lookup(nd, &this, &next, &inode);
1459 if (inode && unlikely(inode->i_op->follow_link) &&
1460 (lookup_flags & LOOKUP_FOLLOW)) {
1461 err = do_follow_link(inode, &next, nd);
1464 nd->inode = nd->path.dentry->d_inode;
1466 path_to_nameidata(&next, nd);
1472 if (lookup_flags & LOOKUP_DIRECTORY) {
1474 if (!nd->inode->i_op->lookup)
1480 nd->last_type = LAST_NORM;
1481 if (this.name[0] != '.')
1484 nd->last_type = LAST_DOT;
1485 else if (this.len == 2 && this.name[1] == '.')
1486 nd->last_type = LAST_DOTDOT;
1491 * We bypassed the ordinary revalidation routines.
1492 * We may need to check the cached dentry for staleness.
1494 if (need_reval_dot(nd->path.dentry)) {
1495 if (nameidata_drop_rcu_last_maybe(nd))
1497 /* Note: we do not d_invalidate() */
1498 err = d_revalidate(nd->path.dentry, nd);
1506 if (nameidata_drop_rcu_last_maybe(nd))
1510 if (!(nd->flags & LOOKUP_RCU))
1511 path_put_conditional(&next, nd);
1514 if (!(nd->flags & LOOKUP_RCU))
1515 path_put(&nd->path);
1520 static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1522 current->total_link_count = 0;
1524 return link_path_walk(name, nd);
1527 static inline int path_walk_simple(const char *name, struct nameidata *nd)
1529 current->total_link_count = 0;
1531 return link_path_walk(name, nd);
1534 static int path_walk(const char *name, struct nameidata *nd)
1536 struct path save = nd->path;
1539 current->total_link_count = 0;
1541 /* make sure the stuff we saved doesn't go away */
1544 result = link_path_walk(name, nd);
1545 if (result == -ESTALE) {
1546 /* nd->path had been dropped */
1547 current->total_link_count = 0;
1549 path_get(&nd->path);
1550 nd->flags |= LOOKUP_REVAL;
1551 result = link_path_walk(name, nd);
1559 static void path_finish_rcu(struct nameidata *nd)
1561 if (nd->flags & LOOKUP_RCU) {
1562 /* RCU dangling. Cancel it. */
1563 nd->flags &= ~LOOKUP_RCU;
1564 nd->root.mnt = NULL;
1566 br_read_unlock(vfsmount_lock);
1572 static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1578 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1579 nd->flags = flags | LOOKUP_RCU;
1581 nd->root.mnt = NULL;
1585 struct fs_struct *fs = current->fs;
1588 br_read_lock(vfsmount_lock);
1592 seq = read_seqcount_begin(&fs->seq);
1593 nd->root = fs->root;
1594 nd->path = nd->root;
1595 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1596 } while (read_seqcount_retry(&fs->seq, seq));
1598 } else if (dfd == AT_FDCWD) {
1599 struct fs_struct *fs = current->fs;
1602 br_read_lock(vfsmount_lock);
1606 seq = read_seqcount_begin(&fs->seq);
1608 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1609 } while (read_seqcount_retry(&fs->seq, seq));
1612 struct dentry *dentry;
1614 file = fget_light(dfd, &fput_needed);
1619 dentry = file->f_path.dentry;
1622 if (!S_ISDIR(dentry->d_inode->i_mode))
1625 retval = file_permission(file, MAY_EXEC);
1629 nd->path = file->f_path;
1633 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1634 br_read_lock(vfsmount_lock);
1637 nd->inode = nd->path.dentry->d_inode;
1641 fput_light(file, fput_needed);
1646 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1652 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1655 nd->root.mnt = NULL;
1659 nd->path = nd->root;
1660 path_get(&nd->root);
1661 } else if (dfd == AT_FDCWD) {
1662 get_fs_pwd(current->fs, &nd->path);
1664 struct dentry *dentry;
1666 file = fget_light(dfd, &fput_needed);
1671 dentry = file->f_path.dentry;
1674 if (!S_ISDIR(dentry->d_inode->i_mode))
1677 retval = file_permission(file, MAY_EXEC);
1681 nd->path = file->f_path;
1682 path_get(&file->f_path);
1684 fput_light(file, fput_needed);
1686 nd->inode = nd->path.dentry->d_inode;
1690 fput_light(file, fput_needed);
1695 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1696 static int do_path_lookup(int dfd, const char *name,
1697 unsigned int flags, struct nameidata *nd)
1702 * Path walking is largely split up into 2 different synchronisation
1703 * schemes, rcu-walk and ref-walk (explained in
1704 * Documentation/filesystems/path-lookup.txt). These share much of the
1705 * path walk code, but some things particularly setup, cleanup, and
1706 * following mounts are sufficiently divergent that functions are
1707 * duplicated. Typically there is a function foo(), and its RCU
1708 * analogue, foo_rcu().
1710 * -ECHILD is the error number of choice (just to avoid clashes) that
1711 * is returned if some aspect of an rcu-walk fails. Such an error must
1712 * be handled by restarting a traditional ref-walk (which will always
1713 * be able to complete).
1715 retval = path_init_rcu(dfd, name, flags, nd);
1716 if (unlikely(retval))
1718 retval = path_walk_rcu(name, nd);
1719 path_finish_rcu(nd);
1721 path_put(&nd->root);
1722 nd->root.mnt = NULL;
1725 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1726 /* slower, locked walk */
1727 if (retval == -ESTALE)
1728 flags |= LOOKUP_REVAL;
1729 retval = path_init(dfd, name, flags, nd);
1730 if (unlikely(retval))
1732 retval = path_walk(name, nd);
1734 path_put(&nd->root);
1735 nd->root.mnt = NULL;
1739 if (likely(!retval)) {
1740 if (unlikely(!audit_dummy_context())) {
1741 if (nd->path.dentry && nd->inode)
1742 audit_inode(name, nd->path.dentry);
1749 int path_lookup(const char *name, unsigned int flags,
1750 struct nameidata *nd)
1752 return do_path_lookup(AT_FDCWD, name, flags, nd);
1755 int kern_path(const char *name, unsigned int flags, struct path *path)
1757 struct nameidata nd;
1758 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1765 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1766 * @dentry: pointer to dentry of the base directory
1767 * @mnt: pointer to vfs mount of the base directory
1768 * @name: pointer to file name
1769 * @flags: lookup flags
1770 * @nd: pointer to nameidata
1772 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1773 const char *name, unsigned int flags,
1774 struct nameidata *nd)
1778 /* same as do_path_lookup */
1779 nd->last_type = LAST_ROOT;
1783 nd->path.dentry = dentry;
1785 path_get(&nd->path);
1786 nd->root = nd->path;
1787 path_get(&nd->root);
1788 nd->inode = nd->path.dentry->d_inode;
1790 retval = path_walk(name, nd);
1791 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1793 audit_inode(name, nd->path.dentry);
1795 path_put(&nd->root);
1796 nd->root.mnt = NULL;
1801 static struct dentry *__lookup_hash(struct qstr *name,
1802 struct dentry *base, struct nameidata *nd)
1804 struct inode *inode = base->d_inode;
1805 struct dentry *dentry;
1808 err = exec_permission(inode, 0);
1810 return ERR_PTR(err);
1813 * See if the low-level filesystem might want
1814 * to use its own hash..
1816 if (base->d_flags & DCACHE_OP_HASH) {
1817 err = base->d_op->d_hash(base, inode, name);
1818 dentry = ERR_PTR(err);
1824 * Don't bother with __d_lookup: callers are for creat as
1825 * well as unlink, so a lot of the time it would cost
1828 dentry = d_lookup(base, name);
1830 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1831 dentry = do_revalidate(dentry, nd);
1834 dentry = d_alloc_and_lookup(base, name, nd);
1840 * Restricted form of lookup. Doesn't follow links, single-component only,
1841 * needs parent already locked. Doesn't follow mounts.
1844 static struct dentry *lookup_hash(struct nameidata *nd)
1846 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1849 static int __lookup_one_len(const char *name, struct qstr *this,
1850 struct dentry *base, int len)
1860 hash = init_name_hash();
1862 c = *(const unsigned char *)name++;
1863 if (c == '/' || c == '\0')
1865 hash = partial_name_hash(c, hash);
1867 this->hash = end_name_hash(hash);
1872 * lookup_one_len - filesystem helper to lookup single pathname component
1873 * @name: pathname component to lookup
1874 * @base: base directory to lookup from
1875 * @len: maximum length @len should be interpreted to
1877 * Note that this routine is purely a helper for filesystem usage and should
1878 * not be called by generic code. Also note that by using this function the
1879 * nameidata argument is passed to the filesystem methods and a filesystem
1880 * using this helper needs to be prepared for that.
1882 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1887 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1889 err = __lookup_one_len(name, &this, base, len);
1891 return ERR_PTR(err);
1893 return __lookup_hash(&this, base, NULL);
1896 int user_path_at(int dfd, const char __user *name, unsigned flags,
1899 struct nameidata nd;
1900 char *tmp = getname(name);
1901 int err = PTR_ERR(tmp);
1904 BUG_ON(flags & LOOKUP_PARENT);
1906 err = do_path_lookup(dfd, tmp, flags, &nd);
1914 static int user_path_parent(int dfd, const char __user *path,
1915 struct nameidata *nd, char **name)
1917 char *s = getname(path);
1923 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1933 * It's inline, so penalty for filesystems that don't use sticky bit is
1936 static inline int check_sticky(struct inode *dir, struct inode *inode)
1938 uid_t fsuid = current_fsuid();
1940 if (!(dir->i_mode & S_ISVTX))
1942 if (inode->i_uid == fsuid)
1944 if (dir->i_uid == fsuid)
1946 return !capable(CAP_FOWNER);
1950 * Check whether we can remove a link victim from directory dir, check
1951 * whether the type of victim is right.
1952 * 1. We can't do it if dir is read-only (done in permission())
1953 * 2. We should have write and exec permissions on dir
1954 * 3. We can't remove anything from append-only dir
1955 * 4. We can't do anything with immutable dir (done in permission())
1956 * 5. If the sticky bit on dir is set we should either
1957 * a. be owner of dir, or
1958 * b. be owner of victim, or
1959 * c. have CAP_FOWNER capability
1960 * 6. If the victim is append-only or immutable we can't do antyhing with
1961 * links pointing to it.
1962 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1963 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1964 * 9. We can't remove a root or mountpoint.
1965 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1966 * nfs_async_unlink().
1968 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1972 if (!victim->d_inode)
1975 BUG_ON(victim->d_parent->d_inode != dir);
1976 audit_inode_child(victim, dir);
1978 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1983 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1984 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1987 if (!S_ISDIR(victim->d_inode->i_mode))
1989 if (IS_ROOT(victim))
1991 } else if (S_ISDIR(victim->d_inode->i_mode))
1993 if (IS_DEADDIR(dir))
1995 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2000 /* Check whether we can create an object with dentry child in directory
2002 * 1. We can't do it if child already exists (open has special treatment for
2003 * this case, but since we are inlined it's OK)
2004 * 2. We can't do it if dir is read-only (done in permission())
2005 * 3. We should have write and exec permissions on dir
2006 * 4. We can't do it if dir is immutable (done in permission())
2008 static inline int may_create(struct inode *dir, struct dentry *child)
2012 if (IS_DEADDIR(dir))
2014 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2018 * p1 and p2 should be directories on the same fs.
2020 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2025 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2029 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2031 p = d_ancestor(p2, p1);
2033 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2034 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2038 p = d_ancestor(p1, p2);
2040 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2041 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2045 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2046 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2050 void unlock_rename(struct dentry *p1, struct dentry *p2)
2052 mutex_unlock(&p1->d_inode->i_mutex);
2054 mutex_unlock(&p2->d_inode->i_mutex);
2055 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2059 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
2060 struct nameidata *nd)
2062 int error = may_create(dir, dentry);
2067 if (!dir->i_op->create)
2068 return -EACCES; /* shouldn't it be ENOSYS? */
2071 error = security_inode_create(dir, dentry, mode);
2074 error = dir->i_op->create(dir, dentry, mode, nd);
2076 fsnotify_create(dir, dentry);
2080 int may_open(struct path *path, int acc_mode, int flag)
2082 struct dentry *dentry = path->dentry;
2083 struct inode *inode = dentry->d_inode;
2089 switch (inode->i_mode & S_IFMT) {
2093 if (acc_mode & MAY_WRITE)
2098 if (path->mnt->mnt_flags & MNT_NODEV)
2107 error = inode_permission(inode, acc_mode);
2112 * An append-only file must be opened in append mode for writing.
2114 if (IS_APPEND(inode)) {
2115 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2121 /* O_NOATIME can only be set by the owner or superuser */
2122 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2126 * Ensure there are no outstanding leases on the file.
2128 return break_lease(inode, flag);
2131 static int handle_truncate(struct file *filp)
2133 struct path *path = &filp->f_path;
2134 struct inode *inode = path->dentry->d_inode;
2135 int error = get_write_access(inode);
2139 * Refuse to truncate files with mandatory locks held on them.
2141 error = locks_verify_locked(inode);
2143 error = security_path_truncate(path);
2145 error = do_truncate(path->dentry, 0,
2146 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2149 put_write_access(inode);
2154 * Be careful about ever adding any more callers of this
2155 * function. Its flags must be in the namei format, not
2156 * what get passed to sys_open().
2158 static int __open_namei_create(struct nameidata *nd, struct path *path,
2159 int open_flag, int mode)
2162 struct dentry *dir = nd->path.dentry;
2164 if (!IS_POSIXACL(dir->d_inode))
2165 mode &= ~current_umask();
2166 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2169 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
2171 mutex_unlock(&dir->d_inode->i_mutex);
2172 dput(nd->path.dentry);
2173 nd->path.dentry = path->dentry;
2177 /* Don't check for write permission, don't truncate */
2178 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
2182 * Note that while the flag value (low two bits) for sys_open means:
2187 * it is changed into
2188 * 00 - no permissions needed
2189 * 01 - read-permission
2190 * 10 - write-permission
2192 * for the internal routines (ie open_namei()/follow_link() etc)
2193 * This is more logical, and also allows the 00 "no perm needed"
2194 * to be used for symlinks (where the permissions are checked
2198 static inline int open_to_namei_flags(int flag)
2200 if ((flag+1) & O_ACCMODE)
2205 static int open_will_truncate(int flag, struct inode *inode)
2208 * We'll never write to the fs underlying
2211 if (special_file(inode->i_mode))
2213 return (flag & O_TRUNC);
2216 static struct file *finish_open(struct nameidata *nd,
2217 int open_flag, int acc_mode)
2223 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
2224 if (will_truncate) {
2225 error = mnt_want_write(nd->path.mnt);
2229 error = may_open(&nd->path, acc_mode, open_flag);
2232 mnt_drop_write(nd->path.mnt);
2235 filp = nameidata_to_filp(nd);
2236 if (!IS_ERR(filp)) {
2237 error = ima_file_check(filp, acc_mode);
2240 filp = ERR_PTR(error);
2243 if (!IS_ERR(filp)) {
2244 if (will_truncate) {
2245 error = handle_truncate(filp);
2248 filp = ERR_PTR(error);
2253 * It is now safe to drop the mnt write
2254 * because the filp has had a write taken
2258 mnt_drop_write(nd->path.mnt);
2259 path_put(&nd->path);
2263 path_put(&nd->path);
2264 return ERR_PTR(error);
2268 * Handle O_CREAT case for do_filp_open
2270 static struct file *do_last(struct nameidata *nd, struct path *path,
2271 int open_flag, int acc_mode,
2272 int mode, const char *pathname)
2274 struct dentry *dir = nd->path.dentry;
2276 int error = -EISDIR;
2278 switch (nd->last_type) {
2281 dir = nd->path.dentry;
2283 if (need_reval_dot(dir)) {
2284 int status = d_revalidate(nd->path.dentry, nd);
2296 audit_inode(pathname, dir);
2300 /* trailing slashes? */
2301 if (nd->last.name[nd->last.len])
2304 mutex_lock(&dir->d_inode->i_mutex);
2306 path->dentry = lookup_hash(nd);
2307 path->mnt = nd->path.mnt;
2309 error = PTR_ERR(path->dentry);
2310 if (IS_ERR(path->dentry)) {
2311 mutex_unlock(&dir->d_inode->i_mutex);
2315 if (IS_ERR(nd->intent.open.file)) {
2316 error = PTR_ERR(nd->intent.open.file);
2317 goto exit_mutex_unlock;
2320 /* Negative dentry, just create the file */
2321 if (!path->dentry->d_inode) {
2323 * This write is needed to ensure that a
2324 * ro->rw transition does not occur between
2325 * the time when the file is created and when
2326 * a permanent write count is taken through
2327 * the 'struct file' in nameidata_to_filp().
2329 error = mnt_want_write(nd->path.mnt);
2331 goto exit_mutex_unlock;
2332 error = __open_namei_create(nd, path, open_flag, mode);
2334 mnt_drop_write(nd->path.mnt);
2337 filp = nameidata_to_filp(nd);
2338 mnt_drop_write(nd->path.mnt);
2339 path_put(&nd->path);
2340 if (!IS_ERR(filp)) {
2341 error = ima_file_check(filp, acc_mode);
2344 filp = ERR_PTR(error);
2351 * It already exists.
2353 mutex_unlock(&dir->d_inode->i_mutex);
2354 audit_inode(pathname, path->dentry);
2357 if (open_flag & O_EXCL)
2360 error = follow_managed(path, nd->flags);
2365 if (!path->dentry->d_inode)
2368 if (path->dentry->d_inode->i_op->follow_link)
2371 path_to_nameidata(path, nd);
2372 nd->inode = path->dentry->d_inode;
2374 if (S_ISDIR(nd->inode->i_mode))
2377 filp = finish_open(nd, open_flag, acc_mode);
2381 mutex_unlock(&dir->d_inode->i_mutex);
2383 path_put_conditional(path, nd);
2385 path_put(&nd->path);
2386 return ERR_PTR(error);
2390 * Note that the low bits of the passed in "open_flag"
2391 * are not the same as in the local variable "flag". See
2392 * open_to_namei_flags() for more details.
2394 struct file *do_filp_open(int dfd, const char *pathname,
2395 int open_flag, int mode, int acc_mode)
2398 struct nameidata nd;
2402 int flag = open_to_namei_flags(open_flag);
2405 if (!(open_flag & O_CREAT))
2408 /* Must never be set by userspace */
2409 open_flag &= ~FMODE_NONOTIFY;
2412 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2413 * check for O_DSYNC if the need any syncing at all we enforce it's
2414 * always set instead of having to deal with possibly weird behaviour
2415 * for malicious applications setting only __O_SYNC.
2417 if (open_flag & __O_SYNC)
2418 open_flag |= O_DSYNC;
2421 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
2423 /* O_TRUNC implies we need access checks for write permissions */
2424 if (open_flag & O_TRUNC)
2425 acc_mode |= MAY_WRITE;
2427 /* Allow the LSM permission hook to distinguish append
2428 access from general write access. */
2429 if (open_flag & O_APPEND)
2430 acc_mode |= MAY_APPEND;
2432 flags = LOOKUP_OPEN;
2433 if (open_flag & O_CREAT) {
2434 flags |= LOOKUP_CREATE;
2435 if (open_flag & O_EXCL)
2436 flags |= LOOKUP_EXCL;
2438 if (open_flag & O_DIRECTORY)
2439 flags |= LOOKUP_DIRECTORY;
2440 if (!(open_flag & O_NOFOLLOW))
2441 flags |= LOOKUP_FOLLOW;
2443 filp = get_empty_filp();
2445 return ERR_PTR(-ENFILE);
2447 filp->f_flags = open_flag;
2448 nd.intent.open.file = filp;
2449 nd.intent.open.flags = flag;
2450 nd.intent.open.create_mode = mode;
2452 if (open_flag & O_CREAT)
2455 /* !O_CREAT, simple open */
2456 error = do_path_lookup(dfd, pathname, flags, &nd);
2457 if (unlikely(error))
2460 if (!(nd.flags & LOOKUP_FOLLOW)) {
2461 if (nd.inode->i_op->follow_link)
2465 if (nd.flags & LOOKUP_DIRECTORY) {
2466 if (!nd.inode->i_op->lookup)
2469 audit_inode(pathname, nd.path.dentry);
2470 filp = finish_open(&nd, open_flag, acc_mode);
2472 release_open_intent(&nd);
2478 filp = ERR_PTR(error);
2482 /* OK, have to create the file. Find the parent. */
2483 error = path_init_rcu(dfd, pathname,
2484 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2487 error = path_walk_rcu(pathname, &nd);
2488 path_finish_rcu(&nd);
2489 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2490 /* slower, locked walk */
2491 if (error == -ESTALE) {
2493 flags |= LOOKUP_REVAL;
2495 error = path_init(dfd, pathname,
2496 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2500 error = path_walk_simple(pathname, &nd);
2502 if (unlikely(error))
2504 if (unlikely(!audit_dummy_context()))
2505 audit_inode(pathname, nd.path.dentry);
2508 * We have the parent and last component.
2511 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2512 while (unlikely(!filp)) { /* trailing symlink */
2513 struct path link = path;
2514 struct inode *linki = link.dentry->d_inode;
2517 if (!(nd.flags & LOOKUP_FOLLOW))
2522 * This is subtle. Instead of calling do_follow_link() we do
2523 * the thing by hands. The reason is that this way we have zero
2524 * link_count and path_walk() (called from ->follow_link)
2525 * honoring LOOKUP_PARENT. After that we have the parent and
2526 * last component, i.e. we are in the same situation as after
2527 * the first path_walk(). Well, almost - if the last component
2528 * is normal we get its copy stored in nd->last.name and we will
2529 * have to putname() it when we are done. Procfs-like symlinks
2530 * just set LAST_BIND.
2532 nd.flags |= LOOKUP_PARENT;
2533 error = security_inode_follow_link(link.dentry, &nd);
2536 error = __do_follow_link(&link, &nd, &cookie);
2537 if (unlikely(error)) {
2538 if (!IS_ERR(cookie) && linki->i_op->put_link)
2539 linki->i_op->put_link(link.dentry, &nd, cookie);
2540 /* nd.path had been dropped */
2544 nd.flags &= ~LOOKUP_PARENT;
2545 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2546 if (linki->i_op->put_link)
2547 linki->i_op->put_link(link.dentry, &nd, cookie);
2553 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
2555 release_open_intent(&nd);
2559 path_put_conditional(&path, &nd);
2563 filp = ERR_PTR(error);
2568 * filp_open - open file and return file pointer
2570 * @filename: path to open
2571 * @flags: open flags as per the open(2) second argument
2572 * @mode: mode for the new file if O_CREAT is set, else ignored
2574 * This is the helper to open a file from kernelspace if you really
2575 * have to. But in generally you should not do this, so please move
2576 * along, nothing to see here..
2578 struct file *filp_open(const char *filename, int flags, int mode)
2580 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
2582 EXPORT_SYMBOL(filp_open);
2585 * lookup_create - lookup a dentry, creating it if it doesn't exist
2586 * @nd: nameidata info
2587 * @is_dir: directory flag
2589 * Simple function to lookup and return a dentry and create it
2590 * if it doesn't exist. Is SMP-safe.
2592 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2594 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2596 struct dentry *dentry = ERR_PTR(-EEXIST);
2598 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2600 * Yucky last component or no last component at all?
2601 * (foo/., foo/.., /////)
2603 if (nd->last_type != LAST_NORM)
2605 nd->flags &= ~LOOKUP_PARENT;
2606 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2607 nd->intent.open.flags = O_EXCL;
2610 * Do the final lookup.
2612 dentry = lookup_hash(nd);
2616 if (dentry->d_inode)
2619 * Special case - lookup gave negative, but... we had foo/bar/
2620 * From the vfs_mknod() POV we just have a negative dentry -
2621 * all is fine. Let's be bastards - you had / on the end, you've
2622 * been asking for (non-existent) directory. -ENOENT for you.
2624 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2626 dentry = ERR_PTR(-ENOENT);
2631 dentry = ERR_PTR(-EEXIST);
2635 EXPORT_SYMBOL_GPL(lookup_create);
2637 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2639 int error = may_create(dir, dentry);
2644 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2647 if (!dir->i_op->mknod)
2650 error = devcgroup_inode_mknod(mode, dev);
2654 error = security_inode_mknod(dir, dentry, mode, dev);
2658 error = dir->i_op->mknod(dir, dentry, mode, dev);
2660 fsnotify_create(dir, dentry);
2664 static int may_mknod(mode_t mode)
2666 switch (mode & S_IFMT) {
2672 case 0: /* zero mode translates to S_IFREG */
2681 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2686 struct dentry *dentry;
2687 struct nameidata nd;
2692 error = user_path_parent(dfd, filename, &nd, &tmp);
2696 dentry = lookup_create(&nd, 0);
2697 if (IS_ERR(dentry)) {
2698 error = PTR_ERR(dentry);
2701 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2702 mode &= ~current_umask();
2703 error = may_mknod(mode);
2706 error = mnt_want_write(nd.path.mnt);
2709 error = security_path_mknod(&nd.path, dentry, mode, dev);
2711 goto out_drop_write;
2712 switch (mode & S_IFMT) {
2713 case 0: case S_IFREG:
2714 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2716 case S_IFCHR: case S_IFBLK:
2717 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2718 new_decode_dev(dev));
2720 case S_IFIFO: case S_IFSOCK:
2721 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2725 mnt_drop_write(nd.path.mnt);
2729 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2736 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2738 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2741 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2743 int error = may_create(dir, dentry);
2748 if (!dir->i_op->mkdir)
2751 mode &= (S_IRWXUGO|S_ISVTX);
2752 error = security_inode_mkdir(dir, dentry, mode);
2756 error = dir->i_op->mkdir(dir, dentry, mode);
2758 fsnotify_mkdir(dir, dentry);
2762 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2766 struct dentry *dentry;
2767 struct nameidata nd;
2769 error = user_path_parent(dfd, pathname, &nd, &tmp);
2773 dentry = lookup_create(&nd, 1);
2774 error = PTR_ERR(dentry);
2778 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2779 mode &= ~current_umask();
2780 error = mnt_want_write(nd.path.mnt);
2783 error = security_path_mkdir(&nd.path, dentry, mode);
2785 goto out_drop_write;
2786 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2788 mnt_drop_write(nd.path.mnt);
2792 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2799 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2801 return sys_mkdirat(AT_FDCWD, pathname, mode);
2805 * We try to drop the dentry early: we should have
2806 * a usage count of 2 if we're the only user of this
2807 * dentry, and if that is true (possibly after pruning
2808 * the dcache), then we drop the dentry now.
2810 * A low-level filesystem can, if it choses, legally
2813 * if (!d_unhashed(dentry))
2816 * if it cannot handle the case of removing a directory
2817 * that is still in use by something else..
2819 void dentry_unhash(struct dentry *dentry)
2822 shrink_dcache_parent(dentry);
2823 spin_lock(&dentry->d_lock);
2824 if (dentry->d_count == 2)
2826 spin_unlock(&dentry->d_lock);
2829 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2831 int error = may_delete(dir, dentry, 1);
2836 if (!dir->i_op->rmdir)
2839 mutex_lock(&dentry->d_inode->i_mutex);
2840 dentry_unhash(dentry);
2841 if (d_mountpoint(dentry))
2844 error = security_inode_rmdir(dir, dentry);
2846 error = dir->i_op->rmdir(dir, dentry);
2848 dentry->d_inode->i_flags |= S_DEAD;
2853 mutex_unlock(&dentry->d_inode->i_mutex);
2862 static long do_rmdir(int dfd, const char __user *pathname)
2866 struct dentry *dentry;
2867 struct nameidata nd;
2869 error = user_path_parent(dfd, pathname, &nd, &name);
2873 switch(nd.last_type) {
2885 nd.flags &= ~LOOKUP_PARENT;
2887 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2888 dentry = lookup_hash(&nd);
2889 error = PTR_ERR(dentry);
2892 error = mnt_want_write(nd.path.mnt);
2895 error = security_path_rmdir(&nd.path, dentry);
2898 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2900 mnt_drop_write(nd.path.mnt);
2904 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2911 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2913 return do_rmdir(AT_FDCWD, pathname);
2916 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2918 int error = may_delete(dir, dentry, 0);
2923 if (!dir->i_op->unlink)
2926 mutex_lock(&dentry->d_inode->i_mutex);
2927 if (d_mountpoint(dentry))
2930 error = security_inode_unlink(dir, dentry);
2932 error = dir->i_op->unlink(dir, dentry);
2937 mutex_unlock(&dentry->d_inode->i_mutex);
2939 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2940 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2941 fsnotify_link_count(dentry->d_inode);
2949 * Make sure that the actual truncation of the file will occur outside its
2950 * directory's i_mutex. Truncate can take a long time if there is a lot of
2951 * writeout happening, and we don't want to prevent access to the directory
2952 * while waiting on the I/O.
2954 static long do_unlinkat(int dfd, const char __user *pathname)
2958 struct dentry *dentry;
2959 struct nameidata nd;
2960 struct inode *inode = NULL;
2962 error = user_path_parent(dfd, pathname, &nd, &name);
2967 if (nd.last_type != LAST_NORM)
2970 nd.flags &= ~LOOKUP_PARENT;
2972 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2973 dentry = lookup_hash(&nd);
2974 error = PTR_ERR(dentry);
2975 if (!IS_ERR(dentry)) {
2976 /* Why not before? Because we want correct error value */
2977 if (nd.last.name[nd.last.len])
2979 inode = dentry->d_inode;
2982 error = mnt_want_write(nd.path.mnt);
2985 error = security_path_unlink(&nd.path, dentry);
2988 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2990 mnt_drop_write(nd.path.mnt);
2994 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2996 iput(inode); /* truncate the inode here */
3003 error = !dentry->d_inode ? -ENOENT :
3004 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3008 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3010 if ((flag & ~AT_REMOVEDIR) != 0)
3013 if (flag & AT_REMOVEDIR)
3014 return do_rmdir(dfd, pathname);
3016 return do_unlinkat(dfd, pathname);
3019 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3021 return do_unlinkat(AT_FDCWD, pathname);
3024 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3026 int error = may_create(dir, dentry);
3031 if (!dir->i_op->symlink)
3034 error = security_inode_symlink(dir, dentry, oldname);
3038 error = dir->i_op->symlink(dir, dentry, oldname);
3040 fsnotify_create(dir, dentry);
3044 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3045 int, newdfd, const char __user *, newname)
3050 struct dentry *dentry;
3051 struct nameidata nd;
3053 from = getname(oldname);
3055 return PTR_ERR(from);
3057 error = user_path_parent(newdfd, newname, &nd, &to);
3061 dentry = lookup_create(&nd, 0);
3062 error = PTR_ERR(dentry);
3066 error = mnt_want_write(nd.path.mnt);
3069 error = security_path_symlink(&nd.path, dentry, from);
3071 goto out_drop_write;
3072 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
3074 mnt_drop_write(nd.path.mnt);
3078 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3086 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3088 return sys_symlinkat(oldname, AT_FDCWD, newname);
3091 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3093 struct inode *inode = old_dentry->d_inode;
3099 error = may_create(dir, new_dentry);
3103 if (dir->i_sb != inode->i_sb)
3107 * A link to an append-only or immutable file cannot be created.
3109 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3111 if (!dir->i_op->link)
3113 if (S_ISDIR(inode->i_mode))
3116 error = security_inode_link(old_dentry, dir, new_dentry);
3120 mutex_lock(&inode->i_mutex);
3121 error = dir->i_op->link(old_dentry, dir, new_dentry);
3122 mutex_unlock(&inode->i_mutex);
3124 fsnotify_link(dir, inode, new_dentry);
3129 * Hardlinks are often used in delicate situations. We avoid
3130 * security-related surprises by not following symlinks on the
3133 * We don't follow them on the oldname either to be compatible
3134 * with linux 2.0, and to avoid hard-linking to directories
3135 * and other special files. --ADM
3137 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3138 int, newdfd, const char __user *, newname, int, flags)
3140 struct dentry *new_dentry;
3141 struct nameidata nd;
3142 struct path old_path;
3146 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
3149 error = user_path_at(olddfd, oldname,
3150 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
3155 error = user_path_parent(newdfd, newname, &nd, &to);
3159 if (old_path.mnt != nd.path.mnt)
3161 new_dentry = lookup_create(&nd, 0);
3162 error = PTR_ERR(new_dentry);
3163 if (IS_ERR(new_dentry))
3165 error = mnt_want_write(nd.path.mnt);
3168 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3170 goto out_drop_write;
3171 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
3173 mnt_drop_write(nd.path.mnt);
3177 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3182 path_put(&old_path);
3187 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3189 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3193 * The worst of all namespace operations - renaming directory. "Perverted"
3194 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3196 * a) we can get into loop creation. Check is done in is_subdir().
3197 * b) race potential - two innocent renames can create a loop together.
3198 * That's where 4.4 screws up. Current fix: serialization on
3199 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3201 * c) we have to lock _three_ objects - parents and victim (if it exists).
3202 * And that - after we got ->i_mutex on parents (until then we don't know
3203 * whether the target exists). Solution: try to be smart with locking
3204 * order for inodes. We rely on the fact that tree topology may change
3205 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3206 * move will be locked. Thus we can rank directories by the tree
3207 * (ancestors first) and rank all non-directories after them.
3208 * That works since everybody except rename does "lock parent, lookup,
3209 * lock child" and rename is under ->s_vfs_rename_mutex.
3210 * HOWEVER, it relies on the assumption that any object with ->lookup()
3211 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3212 * we'd better make sure that there's no link(2) for them.
3213 * d) some filesystems don't support opened-but-unlinked directories,
3214 * either because of layout or because they are not ready to deal with
3215 * all cases correctly. The latter will be fixed (taking this sort of
3216 * stuff into VFS), but the former is not going away. Solution: the same
3217 * trick as in rmdir().
3218 * e) conversion from fhandle to dentry may come in the wrong moment - when
3219 * we are removing the target. Solution: we will have to grab ->i_mutex
3220 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3221 * ->i_mutex on parents, which works but leads to some truly excessive
3224 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3225 struct inode *new_dir, struct dentry *new_dentry)
3228 struct inode *target;
3231 * If we are going to change the parent - check write permissions,
3232 * we'll need to flip '..'.
3234 if (new_dir != old_dir) {
3235 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3240 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3244 target = new_dentry->d_inode;
3246 mutex_lock(&target->i_mutex);
3247 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3251 dentry_unhash(new_dentry);
3252 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3256 target->i_flags |= S_DEAD;
3257 dont_mount(new_dentry);
3259 mutex_unlock(&target->i_mutex);
3260 if (d_unhashed(new_dentry))
3261 d_rehash(new_dentry);
3265 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3266 d_move(old_dentry,new_dentry);
3270 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3271 struct inode *new_dir, struct dentry *new_dentry)
3273 struct inode *target;
3276 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3281 target = new_dentry->d_inode;
3283 mutex_lock(&target->i_mutex);
3284 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3287 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3290 dont_mount(new_dentry);
3291 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3292 d_move(old_dentry, new_dentry);
3295 mutex_unlock(&target->i_mutex);
3300 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3301 struct inode *new_dir, struct dentry *new_dentry)
3304 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3305 const unsigned char *old_name;
3307 if (old_dentry->d_inode == new_dentry->d_inode)
3310 error = may_delete(old_dir, old_dentry, is_dir);
3314 if (!new_dentry->d_inode)
3315 error = may_create(new_dir, new_dentry);
3317 error = may_delete(new_dir, new_dentry, is_dir);
3321 if (!old_dir->i_op->rename)
3324 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3327 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3329 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3331 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3332 new_dentry->d_inode, old_dentry);
3333 fsnotify_oldname_free(old_name);
3338 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3339 int, newdfd, const char __user *, newname)
3341 struct dentry *old_dir, *new_dir;
3342 struct dentry *old_dentry, *new_dentry;
3343 struct dentry *trap;
3344 struct nameidata oldnd, newnd;
3349 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3353 error = user_path_parent(newdfd, newname, &newnd, &to);
3358 if (oldnd.path.mnt != newnd.path.mnt)
3361 old_dir = oldnd.path.dentry;
3363 if (oldnd.last_type != LAST_NORM)
3366 new_dir = newnd.path.dentry;
3367 if (newnd.last_type != LAST_NORM)
3370 oldnd.flags &= ~LOOKUP_PARENT;
3371 newnd.flags &= ~LOOKUP_PARENT;
3372 newnd.flags |= LOOKUP_RENAME_TARGET;
3374 trap = lock_rename(new_dir, old_dir);
3376 old_dentry = lookup_hash(&oldnd);
3377 error = PTR_ERR(old_dentry);
3378 if (IS_ERR(old_dentry))
3380 /* source must exist */
3382 if (!old_dentry->d_inode)
3384 /* unless the source is a directory trailing slashes give -ENOTDIR */
3385 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3387 if (oldnd.last.name[oldnd.last.len])
3389 if (newnd.last.name[newnd.last.len])
3392 /* source should not be ancestor of target */
3394 if (old_dentry == trap)
3396 new_dentry = lookup_hash(&newnd);
3397 error = PTR_ERR(new_dentry);
3398 if (IS_ERR(new_dentry))
3400 /* target should not be an ancestor of source */
3402 if (new_dentry == trap)
3405 error = mnt_want_write(oldnd.path.mnt);
3408 error = security_path_rename(&oldnd.path, old_dentry,
3409 &newnd.path, new_dentry);
3412 error = vfs_rename(old_dir->d_inode, old_dentry,
3413 new_dir->d_inode, new_dentry);
3415 mnt_drop_write(oldnd.path.mnt);
3421 unlock_rename(new_dir, old_dir);
3423 path_put(&newnd.path);
3426 path_put(&oldnd.path);
3432 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3434 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3437 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3441 len = PTR_ERR(link);
3446 if (len > (unsigned) buflen)
3448 if (copy_to_user(buffer, link, len))
3455 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3456 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3457 * using) it for any given inode is up to filesystem.
3459 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3461 struct nameidata nd;
3466 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3468 return PTR_ERR(cookie);
3470 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3471 if (dentry->d_inode->i_op->put_link)
3472 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3476 int vfs_follow_link(struct nameidata *nd, const char *link)
3478 return __vfs_follow_link(nd, link);
3481 /* get the link contents into pagecache */
3482 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3486 struct address_space *mapping = dentry->d_inode->i_mapping;
3487 page = read_mapping_page(mapping, 0, NULL);
3492 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3496 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3498 struct page *page = NULL;
3499 char *s = page_getlink(dentry, &page);
3500 int res = vfs_readlink(dentry,buffer,buflen,s);
3503 page_cache_release(page);
3508 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3510 struct page *page = NULL;
3511 nd_set_link(nd, page_getlink(dentry, &page));
3515 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3517 struct page *page = cookie;
3521 page_cache_release(page);
3526 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3528 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3530 struct address_space *mapping = inode->i_mapping;
3535 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3537 flags |= AOP_FLAG_NOFS;
3540 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3541 flags, &page, &fsdata);
3545 kaddr = kmap_atomic(page, KM_USER0);
3546 memcpy(kaddr, symname, len-1);
3547 kunmap_atomic(kaddr, KM_USER0);
3549 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3556 mark_inode_dirty(inode);
3562 int page_symlink(struct inode *inode, const char *symname, int len)
3564 return __page_symlink(inode, symname, len,
3565 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3568 const struct inode_operations page_symlink_inode_operations = {
3569 .readlink = generic_readlink,
3570 .follow_link = page_follow_link_light,
3571 .put_link = page_put_link,
3574 EXPORT_SYMBOL(user_path_at);
3575 EXPORT_SYMBOL(follow_down_one);
3576 EXPORT_SYMBOL(follow_down);
3577 EXPORT_SYMBOL(follow_up);
3578 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3579 EXPORT_SYMBOL(getname);
3580 EXPORT_SYMBOL(lock_rename);
3581 EXPORT_SYMBOL(lookup_one_len);
3582 EXPORT_SYMBOL(page_follow_link_light);
3583 EXPORT_SYMBOL(page_put_link);
3584 EXPORT_SYMBOL(page_readlink);
3585 EXPORT_SYMBOL(__page_symlink);
3586 EXPORT_SYMBOL(page_symlink);
3587 EXPORT_SYMBOL(page_symlink_inode_operations);
3588 EXPORT_SYMBOL(path_lookup);
3589 EXPORT_SYMBOL(kern_path);
3590 EXPORT_SYMBOL(vfs_path_lookup);
3591 EXPORT_SYMBOL(inode_permission);
3592 EXPORT_SYMBOL(file_permission);
3593 EXPORT_SYMBOL(unlock_rename);
3594 EXPORT_SYMBOL(vfs_create);
3595 EXPORT_SYMBOL(vfs_follow_link);
3596 EXPORT_SYMBOL(vfs_link);
3597 EXPORT_SYMBOL(vfs_mkdir);
3598 EXPORT_SYMBOL(vfs_mknod);
3599 EXPORT_SYMBOL(generic_permission);
3600 EXPORT_SYMBOL(vfs_readlink);
3601 EXPORT_SYMBOL(vfs_rename);
3602 EXPORT_SYMBOL(vfs_rmdir);
3603 EXPORT_SYMBOL(vfs_symlink);
3604 EXPORT_SYMBOL(vfs_unlink);
3605 EXPORT_SYMBOL(dentry_unhash);
3606 EXPORT_SYMBOL(generic_readlink);