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/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
171 * This does basic POSIX ACL permission checking
173 static int acl_permission_check(struct inode *inode, int mask,
174 int (*check_acl)(struct inode *inode, int mask))
176 umode_t mode = inode->i_mode;
178 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
180 if (current_fsuid() == inode->i_uid)
183 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
184 int error = check_acl(inode, mask);
185 if (error != -EAGAIN)
189 if (in_group_p(inode->i_gid))
194 * If the DACs are ok we don't need any capability check.
196 if ((mask & ~mode) == 0)
202 * generic_permission - check for access rights on a Posix-like filesystem
203 * @inode: inode to check access rights for
204 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
205 * @check_acl: optional callback to check for Posix ACLs
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 int generic_permission(struct inode *inode, int mask,
213 int (*check_acl)(struct inode *inode, int mask))
218 * Do the basic POSIX ACL permission checks.
220 ret = acl_permission_check(inode, mask, check_acl);
225 * Read/write DACs are always overridable.
226 * Executable DACs are overridable if at least one exec bit is set.
228 if (!(mask & MAY_EXEC) || execute_ok(inode))
229 if (capable(CAP_DAC_OVERRIDE))
233 * Searching includes executable on directories, else just read.
235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
281 retval = devcgroup_inode_permission(inode, mask);
285 return security_inode_permission(inode,
286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
294 * Used to check for read/write/execute permissions on an already opened
298 * Do not use this function in new code. All access checks should
299 * be done using inode_permission().
301 int file_permission(struct file *file, int mask)
303 return inode_permission(file->f_path.dentry->d_inode, mask);
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
324 int get_write_access(struct inode * inode)
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
337 int deny_write_access(struct file * file)
339 struct inode *inode = file->f_path.dentry->d_inode;
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
353 * path_get - get a reference to a path
354 * @path: path to get the reference to
356 * Given a path increment the reference count to the dentry and the vfsmount.
358 void path_get(struct path *path)
363 EXPORT_SYMBOL(path_get);
366 * path_put - put a reference to a path
367 * @path: path to put the reference to
369 * Given a path decrement the reference count to the dentry and the vfsmount.
371 void path_put(struct path *path)
376 EXPORT_SYMBOL(path_put);
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
382 void release_open_intent(struct nameidata *nd)
384 if (nd->intent.open.file->f_path.dentry == NULL)
385 put_filp(nd->intent.open.file);
387 fput(nd->intent.open.file);
390 static inline struct dentry *
391 do_revalidate(struct dentry *dentry, struct nameidata *nd)
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
402 if (!d_invalidate(dentry)) {
408 dentry = ERR_PTR(status);
415 * Internal lookup() using the new generic dcache.
418 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
420 struct dentry * dentry = __d_lookup(parent, name);
422 /* lockess __d_lookup may fail due to concurrent d_move()
423 * in some unrelated directory, so try with d_lookup
426 dentry = d_lookup(parent, name);
428 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
429 dentry = do_revalidate(dentry, nd);
435 * Short-cut version of permission(), for calling by
436 * path_walk(), when dcache lock is held. Combines parts
437 * of permission() and generic_permission(), and tests ONLY for
438 * MAY_EXEC permission.
440 * If appropriate, check DAC only. If not appropriate, or
441 * short-cut DAC fails, then call permission() to do more
442 * complete permission check.
444 static int exec_permission_lite(struct inode *inode)
448 if (inode->i_op->permission) {
449 ret = inode->i_op->permission(inode, MAY_EXEC);
454 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
458 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
463 return security_inode_permission(inode, MAY_EXEC);
467 * This is called when everything else fails, and we actually have
468 * to go to the low-level filesystem to find out what we should do..
470 * We get the directory semaphore, and after getting that we also
471 * make sure that nobody added the entry to the dcache in the meantime..
474 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
476 struct dentry * result;
477 struct inode *dir = parent->d_inode;
479 mutex_lock(&dir->i_mutex);
481 * First re-do the cached lookup just in case it was created
482 * while we waited for the directory semaphore..
484 * FIXME! This could use version numbering or similar to
485 * avoid unnecessary cache lookups.
487 * The "dcache_lock" is purely to protect the RCU list walker
488 * from concurrent renames at this point (we mustn't get false
489 * negatives from the RCU list walk here, unlike the optimistic
492 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
494 result = d_lookup(parent, name);
496 struct dentry *dentry;
498 /* Don't create child dentry for a dead directory. */
499 result = ERR_PTR(-ENOENT);
503 dentry = d_alloc(parent, name);
504 result = ERR_PTR(-ENOMEM);
506 result = dir->i_op->lookup(dir, dentry, nd);
513 mutex_unlock(&dir->i_mutex);
518 * Uhhuh! Nasty case: the cache was re-populated while
519 * we waited on the semaphore. Need to revalidate.
521 mutex_unlock(&dir->i_mutex);
522 if (result->d_op && result->d_op->d_revalidate) {
523 result = do_revalidate(result, nd);
525 result = ERR_PTR(-ENOENT);
530 static __always_inline void set_root(struct nameidata *nd)
533 struct fs_struct *fs = current->fs;
534 read_lock(&fs->lock);
537 read_unlock(&fs->lock);
541 static int link_path_walk(const char *, struct nameidata *);
543 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
557 res = link_path_walk(link, nd);
558 if (nd->depth || res || nd->last_type!=LAST_NORM)
561 * If it is an iterative symlinks resolution in open_namei() we
562 * have to copy the last component. And all that crap because of
563 * bloody create() on broken symlinks. Furrfu...
566 if (unlikely(!name)) {
570 strcpy(name, nd->last.name);
571 nd->last.name = name;
575 return PTR_ERR(link);
578 static void path_put_conditional(struct path *path, struct nameidata *nd)
581 if (path->mnt != nd->path.mnt)
585 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
587 dput(nd->path.dentry);
588 if (nd->path.mnt != path->mnt)
589 mntput(nd->path.mnt);
590 nd->path.mnt = path->mnt;
591 nd->path.dentry = path->dentry;
594 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
598 struct dentry *dentry = path->dentry;
600 touch_atime(path->mnt, dentry);
601 nd_set_link(nd, NULL);
603 if (path->mnt != nd->path.mnt) {
604 path_to_nameidata(path, nd);
608 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
609 error = PTR_ERR(cookie);
610 if (!IS_ERR(cookie)) {
611 char *s = nd_get_link(nd);
614 error = __vfs_follow_link(nd, s);
615 if (dentry->d_inode->i_op->put_link)
616 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
622 * This limits recursive symlink follows to 8, while
623 * limiting consecutive symlinks to 40.
625 * Without that kind of total limit, nasty chains of consecutive
626 * symlinks can cause almost arbitrarily long lookups.
628 static inline int do_follow_link(struct path *path, struct nameidata *nd)
631 if (current->link_count >= MAX_NESTED_LINKS)
633 if (current->total_link_count >= 40)
635 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
637 err = security_inode_follow_link(path->dentry, nd);
640 current->link_count++;
641 current->total_link_count++;
643 err = __do_follow_link(path, nd);
645 current->link_count--;
649 path_put_conditional(path, nd);
654 int follow_up(struct path *path)
656 struct vfsmount *parent;
657 struct dentry *mountpoint;
658 spin_lock(&vfsmount_lock);
659 parent = path->mnt->mnt_parent;
660 if (parent == path->mnt) {
661 spin_unlock(&vfsmount_lock);
665 mountpoint = dget(path->mnt->mnt_mountpoint);
666 spin_unlock(&vfsmount_lock);
668 path->dentry = mountpoint;
674 /* no need for dcache_lock, as serialization is taken care in
677 static int __follow_mount(struct path *path)
680 while (d_mountpoint(path->dentry)) {
681 struct vfsmount *mounted = lookup_mnt(path);
688 path->dentry = dget(mounted->mnt_root);
694 static void follow_mount(struct path *path)
696 while (d_mountpoint(path->dentry)) {
697 struct vfsmount *mounted = lookup_mnt(path);
703 path->dentry = dget(mounted->mnt_root);
707 /* no need for dcache_lock, as serialization is taken care in
710 int follow_down(struct path *path)
712 struct vfsmount *mounted;
714 mounted = lookup_mnt(path);
719 path->dentry = dget(mounted->mnt_root);
725 static __always_inline void follow_dotdot(struct nameidata *nd)
730 struct vfsmount *parent;
731 struct dentry *old = nd->path.dentry;
733 if (nd->path.dentry == nd->root.dentry &&
734 nd->path.mnt == nd->root.mnt) {
737 spin_lock(&dcache_lock);
738 if (nd->path.dentry != nd->path.mnt->mnt_root) {
739 nd->path.dentry = dget(nd->path.dentry->d_parent);
740 spin_unlock(&dcache_lock);
744 spin_unlock(&dcache_lock);
745 spin_lock(&vfsmount_lock);
746 parent = nd->path.mnt->mnt_parent;
747 if (parent == nd->path.mnt) {
748 spin_unlock(&vfsmount_lock);
752 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
753 spin_unlock(&vfsmount_lock);
755 mntput(nd->path.mnt);
756 nd->path.mnt = parent;
758 follow_mount(&nd->path);
762 * It's more convoluted than I'd like it to be, but... it's still fairly
763 * small and for now I'd prefer to have fast path as straight as possible.
764 * It _is_ time-critical.
766 static int do_lookup(struct nameidata *nd, struct qstr *name,
769 struct vfsmount *mnt = nd->path.mnt;
770 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
774 if (dentry->d_op && dentry->d_op->d_revalidate)
775 goto need_revalidate;
778 path->dentry = dentry;
779 __follow_mount(path);
783 dentry = real_lookup(nd->path.dentry, name, nd);
789 dentry = do_revalidate(dentry, nd);
797 return PTR_ERR(dentry);
802 * This is the basic name resolution function, turning a pathname into
803 * the final dentry. We expect 'base' to be positive and a directory.
805 * Returns 0 and nd will have valid dentry and mnt on success.
806 * Returns error and drops reference to input namei data on failure.
808 static int link_path_walk(const char *name, struct nameidata *nd)
813 unsigned int lookup_flags = nd->flags;
820 inode = nd->path.dentry->d_inode;
822 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
824 /* At this point we know we have a real path component. */
830 nd->flags |= LOOKUP_CONTINUE;
831 err = exec_permission_lite(inode);
836 c = *(const unsigned char *)name;
838 hash = init_name_hash();
841 hash = partial_name_hash(c, hash);
842 c = *(const unsigned char *)name;
843 } while (c && (c != '/'));
844 this.len = name - (const char *) this.name;
845 this.hash = end_name_hash(hash);
847 /* remove trailing slashes? */
850 while (*++name == '/');
852 goto last_with_slashes;
855 * "." and ".." are special - ".." especially so because it has
856 * to be able to know about the current root directory and
857 * parent relationships.
859 if (this.name[0] == '.') switch (this.len) {
863 if (this.name[1] != '.')
866 inode = nd->path.dentry->d_inode;
872 * See if the low-level filesystem might want
873 * to use its own hash..
875 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
876 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
881 /* This does the actual lookups.. */
882 err = do_lookup(nd, &this, &next);
887 inode = next.dentry->d_inode;
891 if (inode->i_op->follow_link) {
892 err = do_follow_link(&next, nd);
896 inode = nd->path.dentry->d_inode;
900 path_to_nameidata(&next, nd);
902 if (!inode->i_op->lookup)
905 /* here ends the main loop */
908 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
910 /* Clear LOOKUP_CONTINUE iff it was previously unset */
911 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
912 if (lookup_flags & LOOKUP_PARENT)
914 if (this.name[0] == '.') switch (this.len) {
918 if (this.name[1] != '.')
921 inode = nd->path.dentry->d_inode;
926 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
927 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
932 err = do_lookup(nd, &this, &next);
935 inode = next.dentry->d_inode;
936 if ((lookup_flags & LOOKUP_FOLLOW)
937 && inode && inode->i_op->follow_link) {
938 err = do_follow_link(&next, nd);
941 inode = nd->path.dentry->d_inode;
943 path_to_nameidata(&next, nd);
947 if (lookup_flags & LOOKUP_DIRECTORY) {
949 if (!inode->i_op->lookup)
955 nd->last_type = LAST_NORM;
956 if (this.name[0] != '.')
959 nd->last_type = LAST_DOT;
960 else if (this.len == 2 && this.name[1] == '.')
961 nd->last_type = LAST_DOTDOT;
966 * We bypassed the ordinary revalidation routines.
967 * We may need to check the cached dentry for staleness.
969 if (nd->path.dentry && nd->path.dentry->d_sb &&
970 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
972 /* Note: we do not d_invalidate() */
973 if (!nd->path.dentry->d_op->d_revalidate(
974 nd->path.dentry, nd))
980 path_put_conditional(&next, nd);
988 static int path_walk(const char *name, struct nameidata *nd)
990 struct path save = nd->path;
993 current->total_link_count = 0;
995 /* make sure the stuff we saved doesn't go away */
998 result = link_path_walk(name, nd);
999 if (result == -ESTALE) {
1000 /* nd->path had been dropped */
1001 current->total_link_count = 0;
1003 path_get(&nd->path);
1004 nd->flags |= LOOKUP_REVAL;
1005 result = link_path_walk(name, nd);
1013 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1019 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1022 nd->root.mnt = NULL;
1026 nd->path = nd->root;
1027 path_get(&nd->root);
1028 } else if (dfd == AT_FDCWD) {
1029 struct fs_struct *fs = current->fs;
1030 read_lock(&fs->lock);
1033 read_unlock(&fs->lock);
1035 struct dentry *dentry;
1037 file = fget_light(dfd, &fput_needed);
1042 dentry = file->f_path.dentry;
1045 if (!S_ISDIR(dentry->d_inode->i_mode))
1048 retval = file_permission(file, MAY_EXEC);
1052 nd->path = file->f_path;
1053 path_get(&file->f_path);
1055 fput_light(file, fput_needed);
1060 fput_light(file, fput_needed);
1065 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1066 static int do_path_lookup(int dfd, const char *name,
1067 unsigned int flags, struct nameidata *nd)
1069 int retval = path_init(dfd, name, flags, nd);
1071 retval = path_walk(name, nd);
1072 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1073 nd->path.dentry->d_inode))
1074 audit_inode(name, nd->path.dentry);
1076 path_put(&nd->root);
1077 nd->root.mnt = NULL;
1082 int path_lookup(const char *name, unsigned int flags,
1083 struct nameidata *nd)
1085 return do_path_lookup(AT_FDCWD, name, flags, nd);
1088 int kern_path(const char *name, unsigned int flags, struct path *path)
1090 struct nameidata nd;
1091 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1098 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1099 * @dentry: pointer to dentry of the base directory
1100 * @mnt: pointer to vfs mount of the base directory
1101 * @name: pointer to file name
1102 * @flags: lookup flags
1103 * @nd: pointer to nameidata
1105 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1106 const char *name, unsigned int flags,
1107 struct nameidata *nd)
1111 /* same as do_path_lookup */
1112 nd->last_type = LAST_ROOT;
1116 nd->path.dentry = dentry;
1118 path_get(&nd->path);
1119 nd->root = nd->path;
1120 path_get(&nd->root);
1122 retval = path_walk(name, nd);
1123 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1124 nd->path.dentry->d_inode))
1125 audit_inode(name, nd->path.dentry);
1127 path_put(&nd->root);
1128 nd->root.mnt = NULL;
1134 * path_lookup_open - lookup a file path with open intent
1135 * @dfd: the directory to use as base, or AT_FDCWD
1136 * @name: pointer to file name
1137 * @lookup_flags: lookup intent flags
1138 * @nd: pointer to nameidata
1139 * @open_flags: open intent flags
1141 static int path_lookup_open(int dfd, const char *name,
1142 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1144 struct file *filp = get_empty_filp();
1149 nd->intent.open.file = filp;
1150 nd->intent.open.flags = open_flags;
1151 nd->intent.open.create_mode = 0;
1152 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1153 if (IS_ERR(nd->intent.open.file)) {
1155 err = PTR_ERR(nd->intent.open.file);
1156 path_put(&nd->path);
1158 } else if (err != 0)
1159 release_open_intent(nd);
1163 static struct dentry *__lookup_hash(struct qstr *name,
1164 struct dentry *base, struct nameidata *nd)
1166 struct dentry *dentry;
1167 struct inode *inode;
1170 inode = base->d_inode;
1173 * See if the low-level filesystem might want
1174 * to use its own hash..
1176 if (base->d_op && base->d_op->d_hash) {
1177 err = base->d_op->d_hash(base, name);
1178 dentry = ERR_PTR(err);
1183 dentry = cached_lookup(base, name, nd);
1187 /* Don't create child dentry for a dead directory. */
1188 dentry = ERR_PTR(-ENOENT);
1189 if (IS_DEADDIR(inode))
1192 new = d_alloc(base, name);
1193 dentry = ERR_PTR(-ENOMEM);
1196 dentry = inode->i_op->lookup(inode, new, nd);
1207 * Restricted form of lookup. Doesn't follow links, single-component only,
1208 * needs parent already locked. Doesn't follow mounts.
1211 static struct dentry *lookup_hash(struct nameidata *nd)
1215 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1217 return ERR_PTR(err);
1218 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1221 static int __lookup_one_len(const char *name, struct qstr *this,
1222 struct dentry *base, int len)
1232 hash = init_name_hash();
1234 c = *(const unsigned char *)name++;
1235 if (c == '/' || c == '\0')
1237 hash = partial_name_hash(c, hash);
1239 this->hash = end_name_hash(hash);
1244 * lookup_one_len - filesystem helper to lookup single pathname component
1245 * @name: pathname component to lookup
1246 * @base: base directory to lookup from
1247 * @len: maximum length @len should be interpreted to
1249 * Note that this routine is purely a helper for filesystem usage and should
1250 * not be called by generic code. Also note that by using this function the
1251 * nameidata argument is passed to the filesystem methods and a filesystem
1252 * using this helper needs to be prepared for that.
1254 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1259 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1261 err = __lookup_one_len(name, &this, base, len);
1263 return ERR_PTR(err);
1265 err = inode_permission(base->d_inode, MAY_EXEC);
1267 return ERR_PTR(err);
1268 return __lookup_hash(&this, base, NULL);
1271 int user_path_at(int dfd, const char __user *name, unsigned flags,
1274 struct nameidata nd;
1275 char *tmp = getname(name);
1276 int err = PTR_ERR(tmp);
1279 BUG_ON(flags & LOOKUP_PARENT);
1281 err = do_path_lookup(dfd, tmp, flags, &nd);
1289 static int user_path_parent(int dfd, const char __user *path,
1290 struct nameidata *nd, char **name)
1292 char *s = getname(path);
1298 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1308 * It's inline, so penalty for filesystems that don't use sticky bit is
1311 static inline int check_sticky(struct inode *dir, struct inode *inode)
1313 uid_t fsuid = current_fsuid();
1315 if (!(dir->i_mode & S_ISVTX))
1317 if (inode->i_uid == fsuid)
1319 if (dir->i_uid == fsuid)
1321 return !capable(CAP_FOWNER);
1325 * Check whether we can remove a link victim from directory dir, check
1326 * whether the type of victim is right.
1327 * 1. We can't do it if dir is read-only (done in permission())
1328 * 2. We should have write and exec permissions on dir
1329 * 3. We can't remove anything from append-only dir
1330 * 4. We can't do anything with immutable dir (done in permission())
1331 * 5. If the sticky bit on dir is set we should either
1332 * a. be owner of dir, or
1333 * b. be owner of victim, or
1334 * c. have CAP_FOWNER capability
1335 * 6. If the victim is append-only or immutable we can't do antyhing with
1336 * links pointing to it.
1337 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1338 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1339 * 9. We can't remove a root or mountpoint.
1340 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1341 * nfs_async_unlink().
1343 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1347 if (!victim->d_inode)
1350 BUG_ON(victim->d_parent->d_inode != dir);
1351 audit_inode_child(victim->d_name.name, victim, dir);
1353 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1358 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1359 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1362 if (!S_ISDIR(victim->d_inode->i_mode))
1364 if (IS_ROOT(victim))
1366 } else if (S_ISDIR(victim->d_inode->i_mode))
1368 if (IS_DEADDIR(dir))
1370 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1375 /* Check whether we can create an object with dentry child in directory
1377 * 1. We can't do it if child already exists (open has special treatment for
1378 * this case, but since we are inlined it's OK)
1379 * 2. We can't do it if dir is read-only (done in permission())
1380 * 3. We should have write and exec permissions on dir
1381 * 4. We can't do it if dir is immutable (done in permission())
1383 static inline int may_create(struct inode *dir, struct dentry *child)
1387 if (IS_DEADDIR(dir))
1389 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1393 * O_DIRECTORY translates into forcing a directory lookup.
1395 static inline int lookup_flags(unsigned int f)
1397 unsigned long retval = LOOKUP_FOLLOW;
1400 retval &= ~LOOKUP_FOLLOW;
1402 if (f & O_DIRECTORY)
1403 retval |= LOOKUP_DIRECTORY;
1409 * p1 and p2 should be directories on the same fs.
1411 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1416 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1420 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1422 p = d_ancestor(p2, p1);
1424 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1425 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1429 p = d_ancestor(p1, p2);
1431 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1432 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1436 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1437 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1441 void unlock_rename(struct dentry *p1, struct dentry *p2)
1443 mutex_unlock(&p1->d_inode->i_mutex);
1445 mutex_unlock(&p2->d_inode->i_mutex);
1446 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1450 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1451 struct nameidata *nd)
1453 int error = may_create(dir, dentry);
1458 if (!dir->i_op->create)
1459 return -EACCES; /* shouldn't it be ENOSYS? */
1462 error = security_inode_create(dir, dentry, mode);
1466 error = dir->i_op->create(dir, dentry, mode, nd);
1468 fsnotify_create(dir, dentry);
1472 int may_open(struct path *path, int acc_mode, int flag)
1474 struct dentry *dentry = path->dentry;
1475 struct inode *inode = dentry->d_inode;
1481 switch (inode->i_mode & S_IFMT) {
1485 if (acc_mode & MAY_WRITE)
1490 if (path->mnt->mnt_flags & MNT_NODEV)
1499 error = inode_permission(inode, acc_mode);
1503 error = ima_path_check(path, acc_mode ?
1504 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1505 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1511 * An append-only file must be opened in append mode for writing.
1513 if (IS_APPEND(inode)) {
1515 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1521 /* O_NOATIME can only be set by the owner or superuser */
1522 if (flag & O_NOATIME)
1523 if (!is_owner_or_cap(inode)) {
1529 * Ensure there are no outstanding leases on the file.
1531 error = break_lease(inode, flag);
1535 if (flag & O_TRUNC) {
1536 error = get_write_access(inode);
1541 * Refuse to truncate files with mandatory locks held on them.
1543 error = locks_verify_locked(inode);
1545 error = security_path_truncate(path, 0,
1546 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1550 error = do_truncate(dentry, 0,
1551 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1554 put_write_access(inode);
1558 if (flag & FMODE_WRITE)
1563 ima_counts_put(path, acc_mode ?
1564 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1565 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1570 * Be careful about ever adding any more callers of this
1571 * function. Its flags must be in the namei format, not
1572 * what get passed to sys_open().
1574 static int __open_namei_create(struct nameidata *nd, struct path *path,
1578 struct dentry *dir = nd->path.dentry;
1580 if (!IS_POSIXACL(dir->d_inode))
1581 mode &= ~current_umask();
1582 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1585 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1587 mutex_unlock(&dir->d_inode->i_mutex);
1588 dput(nd->path.dentry);
1589 nd->path.dentry = path->dentry;
1592 /* Don't check for write permission, don't truncate */
1593 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1597 * Note that while the flag value (low two bits) for sys_open means:
1602 * it is changed into
1603 * 00 - no permissions needed
1604 * 01 - read-permission
1605 * 10 - write-permission
1607 * for the internal routines (ie open_namei()/follow_link() etc)
1608 * This is more logical, and also allows the 00 "no perm needed"
1609 * to be used for symlinks (where the permissions are checked
1613 static inline int open_to_namei_flags(int flag)
1615 if ((flag+1) & O_ACCMODE)
1620 static int open_will_write_to_fs(int flag, struct inode *inode)
1623 * We'll never write to the fs underlying
1626 if (special_file(inode->i_mode))
1628 return (flag & O_TRUNC);
1632 * Note that the low bits of the passed in "open_flag"
1633 * are not the same as in the local variable "flag". See
1634 * open_to_namei_flags() for more details.
1636 struct file *do_filp_open(int dfd, const char *pathname,
1637 int open_flag, int mode, int acc_mode)
1640 struct nameidata nd;
1642 struct path path, save;
1646 int flag = open_to_namei_flags(open_flag);
1649 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1650 * check for O_DSYNC if the need any syncing at all we enforce it's
1651 * always set instead of having to deal with possibly weird behaviour
1652 * for malicious applications setting only __O_SYNC.
1654 if (open_flag & __O_SYNC)
1655 open_flag |= O_DSYNC;
1658 acc_mode = MAY_OPEN | ACC_MODE(flag);
1660 /* O_TRUNC implies we need access checks for write permissions */
1662 acc_mode |= MAY_WRITE;
1664 /* Allow the LSM permission hook to distinguish append
1665 access from general write access. */
1666 if (flag & O_APPEND)
1667 acc_mode |= MAY_APPEND;
1670 * The simplest case - just a plain lookup.
1672 if (!(flag & O_CREAT)) {
1673 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1676 return ERR_PTR(error);
1681 * Create - we need to know the parent.
1683 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1685 return ERR_PTR(error);
1686 error = path_walk(pathname, &nd);
1690 return ERR_PTR(error);
1692 if (unlikely(!audit_dummy_context()))
1693 audit_inode(pathname, nd.path.dentry);
1696 * We have the parent and last component. First of all, check
1697 * that we are not asked to creat(2) an obvious directory - that
1701 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1705 filp = get_empty_filp();
1708 nd.intent.open.file = filp;
1709 nd.intent.open.flags = flag;
1710 nd.intent.open.create_mode = mode;
1711 dir = nd.path.dentry;
1712 nd.flags &= ~LOOKUP_PARENT;
1713 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1715 nd.flags |= LOOKUP_EXCL;
1716 mutex_lock(&dir->d_inode->i_mutex);
1717 path.dentry = lookup_hash(&nd);
1718 path.mnt = nd.path.mnt;
1721 error = PTR_ERR(path.dentry);
1722 if (IS_ERR(path.dentry)) {
1723 mutex_unlock(&dir->d_inode->i_mutex);
1727 if (IS_ERR(nd.intent.open.file)) {
1728 error = PTR_ERR(nd.intent.open.file);
1729 goto exit_mutex_unlock;
1732 /* Negative dentry, just create the file */
1733 if (!path.dentry->d_inode) {
1735 * This write is needed to ensure that a
1736 * ro->rw transition does not occur between
1737 * the time when the file is created and when
1738 * a permanent write count is taken through
1739 * the 'struct file' in nameidata_to_filp().
1741 error = mnt_want_write(nd.path.mnt);
1743 goto exit_mutex_unlock;
1744 error = __open_namei_create(&nd, &path, flag, mode);
1746 mnt_drop_write(nd.path.mnt);
1749 filp = nameidata_to_filp(&nd, open_flag);
1751 ima_counts_put(&nd.path,
1752 acc_mode & (MAY_READ | MAY_WRITE |
1754 mnt_drop_write(nd.path.mnt);
1761 * It already exists.
1763 mutex_unlock(&dir->d_inode->i_mutex);
1764 audit_inode(pathname, path.dentry);
1770 if (__follow_mount(&path)) {
1772 if (flag & O_NOFOLLOW)
1777 if (!path.dentry->d_inode)
1779 if (path.dentry->d_inode->i_op->follow_link)
1782 path_to_nameidata(&path, &nd);
1784 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1789 * 1. may_open() truncates a file
1790 * 2. a rw->ro mount transition occurs
1791 * 3. nameidata_to_filp() fails due to
1793 * That would be inconsistent, and should
1794 * be avoided. Taking this mnt write here
1795 * ensures that (2) can not occur.
1797 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1799 error = mnt_want_write(nd.path.mnt);
1803 error = may_open(&nd.path, acc_mode, flag);
1806 mnt_drop_write(nd.path.mnt);
1809 filp = nameidata_to_filp(&nd, open_flag);
1811 ima_counts_put(&nd.path,
1812 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1814 * It is now safe to drop the mnt write
1815 * because the filp has had a write taken
1819 mnt_drop_write(nd.path.mnt);
1825 mutex_unlock(&dir->d_inode->i_mutex);
1827 path_put_conditional(&path, &nd);
1829 if (!IS_ERR(nd.intent.open.file))
1830 release_open_intent(&nd);
1835 return ERR_PTR(error);
1839 if (flag & O_NOFOLLOW)
1842 * This is subtle. Instead of calling do_follow_link() we do the
1843 * thing by hands. The reason is that this way we have zero link_count
1844 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1845 * After that we have the parent and last component, i.e.
1846 * we are in the same situation as after the first path_walk().
1847 * Well, almost - if the last component is normal we get its copy
1848 * stored in nd->last.name and we will have to putname() it when we
1849 * are done. Procfs-like symlinks just set LAST_BIND.
1851 nd.flags |= LOOKUP_PARENT;
1852 error = security_inode_follow_link(path.dentry, &nd);
1857 error = __do_follow_link(&path, &nd);
1858 if (error == -ESTALE) {
1859 /* nd.path had been dropped */
1862 nd.flags |= LOOKUP_REVAL;
1863 error = __do_follow_link(&path, &nd);
1868 /* Does someone understand code flow here? Or it is only
1869 * me so stupid? Anathema to whoever designed this non-sense
1870 * with "intent.open".
1872 release_open_intent(&nd);
1875 return ERR_PTR(error);
1877 nd.flags &= ~LOOKUP_PARENT;
1878 if (nd.last_type == LAST_BIND)
1881 if (nd.last_type != LAST_NORM)
1883 if (nd.last.name[nd.last.len]) {
1884 __putname(nd.last.name);
1889 __putname(nd.last.name);
1892 dir = nd.path.dentry;
1893 mutex_lock(&dir->d_inode->i_mutex);
1894 path.dentry = lookup_hash(&nd);
1895 path.mnt = nd.path.mnt;
1896 __putname(nd.last.name);
1901 * filp_open - open file and return file pointer
1903 * @filename: path to open
1904 * @flags: open flags as per the open(2) second argument
1905 * @mode: mode for the new file if O_CREAT is set, else ignored
1907 * This is the helper to open a file from kernelspace if you really
1908 * have to. But in generally you should not do this, so please move
1909 * along, nothing to see here..
1911 struct file *filp_open(const char *filename, int flags, int mode)
1913 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1915 EXPORT_SYMBOL(filp_open);
1918 * lookup_create - lookup a dentry, creating it if it doesn't exist
1919 * @nd: nameidata info
1920 * @is_dir: directory flag
1922 * Simple function to lookup and return a dentry and create it
1923 * if it doesn't exist. Is SMP-safe.
1925 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1927 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1929 struct dentry *dentry = ERR_PTR(-EEXIST);
1931 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1933 * Yucky last component or no last component at all?
1934 * (foo/., foo/.., /////)
1936 if (nd->last_type != LAST_NORM)
1938 nd->flags &= ~LOOKUP_PARENT;
1939 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1940 nd->intent.open.flags = O_EXCL;
1943 * Do the final lookup.
1945 dentry = lookup_hash(nd);
1949 if (dentry->d_inode)
1952 * Special case - lookup gave negative, but... we had foo/bar/
1953 * From the vfs_mknod() POV we just have a negative dentry -
1954 * all is fine. Let's be bastards - you had / on the end, you've
1955 * been asking for (non-existent) directory. -ENOENT for you.
1957 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1959 dentry = ERR_PTR(-ENOENT);
1964 dentry = ERR_PTR(-EEXIST);
1968 EXPORT_SYMBOL_GPL(lookup_create);
1970 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1972 int error = may_create(dir, dentry);
1977 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1980 if (!dir->i_op->mknod)
1983 error = devcgroup_inode_mknod(mode, dev);
1987 error = security_inode_mknod(dir, dentry, mode, dev);
1992 error = dir->i_op->mknod(dir, dentry, mode, dev);
1994 fsnotify_create(dir, dentry);
1998 static int may_mknod(mode_t mode)
2000 switch (mode & S_IFMT) {
2006 case 0: /* zero mode translates to S_IFREG */
2015 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2020 struct dentry *dentry;
2021 struct nameidata nd;
2026 error = user_path_parent(dfd, filename, &nd, &tmp);
2030 dentry = lookup_create(&nd, 0);
2031 if (IS_ERR(dentry)) {
2032 error = PTR_ERR(dentry);
2035 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2036 mode &= ~current_umask();
2037 error = may_mknod(mode);
2040 error = mnt_want_write(nd.path.mnt);
2043 error = security_path_mknod(&nd.path, dentry, mode, dev);
2045 goto out_drop_write;
2046 switch (mode & S_IFMT) {
2047 case 0: case S_IFREG:
2048 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2050 case S_IFCHR: case S_IFBLK:
2051 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2052 new_decode_dev(dev));
2054 case S_IFIFO: case S_IFSOCK:
2055 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2059 mnt_drop_write(nd.path.mnt);
2063 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2070 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2072 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2075 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2077 int error = may_create(dir, dentry);
2082 if (!dir->i_op->mkdir)
2085 mode &= (S_IRWXUGO|S_ISVTX);
2086 error = security_inode_mkdir(dir, dentry, mode);
2091 error = dir->i_op->mkdir(dir, dentry, mode);
2093 fsnotify_mkdir(dir, dentry);
2097 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2101 struct dentry *dentry;
2102 struct nameidata nd;
2104 error = user_path_parent(dfd, pathname, &nd, &tmp);
2108 dentry = lookup_create(&nd, 1);
2109 error = PTR_ERR(dentry);
2113 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2114 mode &= ~current_umask();
2115 error = mnt_want_write(nd.path.mnt);
2118 error = security_path_mkdir(&nd.path, dentry, mode);
2120 goto out_drop_write;
2121 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2123 mnt_drop_write(nd.path.mnt);
2127 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2134 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2136 return sys_mkdirat(AT_FDCWD, pathname, mode);
2140 * We try to drop the dentry early: we should have
2141 * a usage count of 2 if we're the only user of this
2142 * dentry, and if that is true (possibly after pruning
2143 * the dcache), then we drop the dentry now.
2145 * A low-level filesystem can, if it choses, legally
2148 * if (!d_unhashed(dentry))
2151 * if it cannot handle the case of removing a directory
2152 * that is still in use by something else..
2154 void dentry_unhash(struct dentry *dentry)
2157 shrink_dcache_parent(dentry);
2158 spin_lock(&dcache_lock);
2159 spin_lock(&dentry->d_lock);
2160 if (atomic_read(&dentry->d_count) == 2)
2162 spin_unlock(&dentry->d_lock);
2163 spin_unlock(&dcache_lock);
2166 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2168 int error = may_delete(dir, dentry, 1);
2173 if (!dir->i_op->rmdir)
2178 mutex_lock(&dentry->d_inode->i_mutex);
2179 dentry_unhash(dentry);
2180 if (d_mountpoint(dentry))
2183 error = security_inode_rmdir(dir, dentry);
2185 error = dir->i_op->rmdir(dir, dentry);
2187 dentry->d_inode->i_flags |= S_DEAD;
2190 mutex_unlock(&dentry->d_inode->i_mutex);
2199 static long do_rmdir(int dfd, const char __user *pathname)
2203 struct dentry *dentry;
2204 struct nameidata nd;
2206 error = user_path_parent(dfd, pathname, &nd, &name);
2210 switch(nd.last_type) {
2222 nd.flags &= ~LOOKUP_PARENT;
2224 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2225 dentry = lookup_hash(&nd);
2226 error = PTR_ERR(dentry);
2229 error = mnt_want_write(nd.path.mnt);
2232 error = security_path_rmdir(&nd.path, dentry);
2235 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2237 mnt_drop_write(nd.path.mnt);
2241 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2248 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2250 return do_rmdir(AT_FDCWD, pathname);
2253 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2255 int error = may_delete(dir, dentry, 0);
2260 if (!dir->i_op->unlink)
2265 mutex_lock(&dentry->d_inode->i_mutex);
2266 if (d_mountpoint(dentry))
2269 error = security_inode_unlink(dir, dentry);
2271 error = dir->i_op->unlink(dir, dentry);
2273 mutex_unlock(&dentry->d_inode->i_mutex);
2275 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2276 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2277 fsnotify_link_count(dentry->d_inode);
2285 * Make sure that the actual truncation of the file will occur outside its
2286 * directory's i_mutex. Truncate can take a long time if there is a lot of
2287 * writeout happening, and we don't want to prevent access to the directory
2288 * while waiting on the I/O.
2290 static long do_unlinkat(int dfd, const char __user *pathname)
2294 struct dentry *dentry;
2295 struct nameidata nd;
2296 struct inode *inode = NULL;
2298 error = user_path_parent(dfd, pathname, &nd, &name);
2303 if (nd.last_type != LAST_NORM)
2306 nd.flags &= ~LOOKUP_PARENT;
2308 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2309 dentry = lookup_hash(&nd);
2310 error = PTR_ERR(dentry);
2311 if (!IS_ERR(dentry)) {
2312 /* Why not before? Because we want correct error value */
2313 if (nd.last.name[nd.last.len])
2315 inode = dentry->d_inode;
2317 atomic_inc(&inode->i_count);
2318 error = mnt_want_write(nd.path.mnt);
2321 error = security_path_unlink(&nd.path, dentry);
2324 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2326 mnt_drop_write(nd.path.mnt);
2330 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2332 iput(inode); /* truncate the inode here */
2339 error = !dentry->d_inode ? -ENOENT :
2340 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2344 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2346 if ((flag & ~AT_REMOVEDIR) != 0)
2349 if (flag & AT_REMOVEDIR)
2350 return do_rmdir(dfd, pathname);
2352 return do_unlinkat(dfd, pathname);
2355 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2357 return do_unlinkat(AT_FDCWD, pathname);
2360 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2362 int error = may_create(dir, dentry);
2367 if (!dir->i_op->symlink)
2370 error = security_inode_symlink(dir, dentry, oldname);
2375 error = dir->i_op->symlink(dir, dentry, oldname);
2377 fsnotify_create(dir, dentry);
2381 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2382 int, newdfd, const char __user *, newname)
2387 struct dentry *dentry;
2388 struct nameidata nd;
2390 from = getname(oldname);
2392 return PTR_ERR(from);
2394 error = user_path_parent(newdfd, newname, &nd, &to);
2398 dentry = lookup_create(&nd, 0);
2399 error = PTR_ERR(dentry);
2403 error = mnt_want_write(nd.path.mnt);
2406 error = security_path_symlink(&nd.path, dentry, from);
2408 goto out_drop_write;
2409 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2411 mnt_drop_write(nd.path.mnt);
2415 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2423 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2425 return sys_symlinkat(oldname, AT_FDCWD, newname);
2428 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2430 struct inode *inode = old_dentry->d_inode;
2436 error = may_create(dir, new_dentry);
2440 if (dir->i_sb != inode->i_sb)
2444 * A link to an append-only or immutable file cannot be created.
2446 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2448 if (!dir->i_op->link)
2450 if (S_ISDIR(inode->i_mode))
2453 error = security_inode_link(old_dentry, dir, new_dentry);
2457 mutex_lock(&inode->i_mutex);
2459 error = dir->i_op->link(old_dentry, dir, new_dentry);
2460 mutex_unlock(&inode->i_mutex);
2462 fsnotify_link(dir, inode, new_dentry);
2467 * Hardlinks are often used in delicate situations. We avoid
2468 * security-related surprises by not following symlinks on the
2471 * We don't follow them on the oldname either to be compatible
2472 * with linux 2.0, and to avoid hard-linking to directories
2473 * and other special files. --ADM
2475 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2476 int, newdfd, const char __user *, newname, int, flags)
2478 struct dentry *new_dentry;
2479 struct nameidata nd;
2480 struct path old_path;
2484 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2487 error = user_path_at(olddfd, oldname,
2488 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2493 error = user_path_parent(newdfd, newname, &nd, &to);
2497 if (old_path.mnt != nd.path.mnt)
2499 new_dentry = lookup_create(&nd, 0);
2500 error = PTR_ERR(new_dentry);
2501 if (IS_ERR(new_dentry))
2503 error = mnt_want_write(nd.path.mnt);
2506 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2508 goto out_drop_write;
2509 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2511 mnt_drop_write(nd.path.mnt);
2515 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2520 path_put(&old_path);
2525 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2527 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2531 * The worst of all namespace operations - renaming directory. "Perverted"
2532 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2534 * a) we can get into loop creation. Check is done in is_subdir().
2535 * b) race potential - two innocent renames can create a loop together.
2536 * That's where 4.4 screws up. Current fix: serialization on
2537 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2539 * c) we have to lock _three_ objects - parents and victim (if it exists).
2540 * And that - after we got ->i_mutex on parents (until then we don't know
2541 * whether the target exists). Solution: try to be smart with locking
2542 * order for inodes. We rely on the fact that tree topology may change
2543 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2544 * move will be locked. Thus we can rank directories by the tree
2545 * (ancestors first) and rank all non-directories after them.
2546 * That works since everybody except rename does "lock parent, lookup,
2547 * lock child" and rename is under ->s_vfs_rename_mutex.
2548 * HOWEVER, it relies on the assumption that any object with ->lookup()
2549 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2550 * we'd better make sure that there's no link(2) for them.
2551 * d) some filesystems don't support opened-but-unlinked directories,
2552 * either because of layout or because they are not ready to deal with
2553 * all cases correctly. The latter will be fixed (taking this sort of
2554 * stuff into VFS), but the former is not going away. Solution: the same
2555 * trick as in rmdir().
2556 * e) conversion from fhandle to dentry may come in the wrong moment - when
2557 * we are removing the target. Solution: we will have to grab ->i_mutex
2558 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2559 * ->i_mutex on parents, which works but leads to some truely excessive
2562 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2563 struct inode *new_dir, struct dentry *new_dentry)
2566 struct inode *target;
2569 * If we are going to change the parent - check write permissions,
2570 * we'll need to flip '..'.
2572 if (new_dir != old_dir) {
2573 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2578 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2582 target = new_dentry->d_inode;
2584 mutex_lock(&target->i_mutex);
2585 dentry_unhash(new_dentry);
2587 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2590 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2593 target->i_flags |= S_DEAD;
2594 mutex_unlock(&target->i_mutex);
2595 if (d_unhashed(new_dentry))
2596 d_rehash(new_dentry);
2600 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2601 d_move(old_dentry,new_dentry);
2605 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2606 struct inode *new_dir, struct dentry *new_dentry)
2608 struct inode *target;
2611 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2616 target = new_dentry->d_inode;
2618 mutex_lock(&target->i_mutex);
2619 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2622 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2625 d_move(old_dentry, new_dentry);
2628 mutex_unlock(&target->i_mutex);
2633 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2634 struct inode *new_dir, struct dentry *new_dentry)
2637 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2638 const char *old_name;
2640 if (old_dentry->d_inode == new_dentry->d_inode)
2643 error = may_delete(old_dir, old_dentry, is_dir);
2647 if (!new_dentry->d_inode)
2648 error = may_create(new_dir, new_dentry);
2650 error = may_delete(new_dir, new_dentry, is_dir);
2654 if (!old_dir->i_op->rename)
2657 vfs_dq_init(old_dir);
2658 vfs_dq_init(new_dir);
2660 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2663 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2665 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2667 const char *new_name = old_dentry->d_name.name;
2668 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2669 new_dentry->d_inode, old_dentry);
2671 fsnotify_oldname_free(old_name);
2676 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2677 int, newdfd, const char __user *, newname)
2679 struct dentry *old_dir, *new_dir;
2680 struct dentry *old_dentry, *new_dentry;
2681 struct dentry *trap;
2682 struct nameidata oldnd, newnd;
2687 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2691 error = user_path_parent(newdfd, newname, &newnd, &to);
2696 if (oldnd.path.mnt != newnd.path.mnt)
2699 old_dir = oldnd.path.dentry;
2701 if (oldnd.last_type != LAST_NORM)
2704 new_dir = newnd.path.dentry;
2705 if (newnd.last_type != LAST_NORM)
2708 oldnd.flags &= ~LOOKUP_PARENT;
2709 newnd.flags &= ~LOOKUP_PARENT;
2710 newnd.flags |= LOOKUP_RENAME_TARGET;
2712 trap = lock_rename(new_dir, old_dir);
2714 old_dentry = lookup_hash(&oldnd);
2715 error = PTR_ERR(old_dentry);
2716 if (IS_ERR(old_dentry))
2718 /* source must exist */
2720 if (!old_dentry->d_inode)
2722 /* unless the source is a directory trailing slashes give -ENOTDIR */
2723 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2725 if (oldnd.last.name[oldnd.last.len])
2727 if (newnd.last.name[newnd.last.len])
2730 /* source should not be ancestor of target */
2732 if (old_dentry == trap)
2734 new_dentry = lookup_hash(&newnd);
2735 error = PTR_ERR(new_dentry);
2736 if (IS_ERR(new_dentry))
2738 /* target should not be an ancestor of source */
2740 if (new_dentry == trap)
2743 error = mnt_want_write(oldnd.path.mnt);
2746 error = security_path_rename(&oldnd.path, old_dentry,
2747 &newnd.path, new_dentry);
2750 error = vfs_rename(old_dir->d_inode, old_dentry,
2751 new_dir->d_inode, new_dentry);
2753 mnt_drop_write(oldnd.path.mnt);
2759 unlock_rename(new_dir, old_dir);
2761 path_put(&newnd.path);
2764 path_put(&oldnd.path);
2770 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2772 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2775 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2779 len = PTR_ERR(link);
2784 if (len > (unsigned) buflen)
2786 if (copy_to_user(buffer, link, len))
2793 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2794 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2795 * using) it for any given inode is up to filesystem.
2797 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2799 struct nameidata nd;
2804 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2806 return PTR_ERR(cookie);
2808 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2809 if (dentry->d_inode->i_op->put_link)
2810 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2814 int vfs_follow_link(struct nameidata *nd, const char *link)
2816 return __vfs_follow_link(nd, link);
2819 /* get the link contents into pagecache */
2820 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2824 struct address_space *mapping = dentry->d_inode->i_mapping;
2825 page = read_mapping_page(mapping, 0, NULL);
2830 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2834 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2836 struct page *page = NULL;
2837 char *s = page_getlink(dentry, &page);
2838 int res = vfs_readlink(dentry,buffer,buflen,s);
2841 page_cache_release(page);
2846 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2848 struct page *page = NULL;
2849 nd_set_link(nd, page_getlink(dentry, &page));
2853 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2855 struct page *page = cookie;
2859 page_cache_release(page);
2864 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2866 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2868 struct address_space *mapping = inode->i_mapping;
2873 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2875 flags |= AOP_FLAG_NOFS;
2878 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2879 flags, &page, &fsdata);
2883 kaddr = kmap_atomic(page, KM_USER0);
2884 memcpy(kaddr, symname, len-1);
2885 kunmap_atomic(kaddr, KM_USER0);
2887 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2894 mark_inode_dirty(inode);
2900 int page_symlink(struct inode *inode, const char *symname, int len)
2902 return __page_symlink(inode, symname, len,
2903 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2906 const struct inode_operations page_symlink_inode_operations = {
2907 .readlink = generic_readlink,
2908 .follow_link = page_follow_link_light,
2909 .put_link = page_put_link,
2912 EXPORT_SYMBOL(user_path_at);
2913 EXPORT_SYMBOL(follow_down);
2914 EXPORT_SYMBOL(follow_up);
2915 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2916 EXPORT_SYMBOL(getname);
2917 EXPORT_SYMBOL(lock_rename);
2918 EXPORT_SYMBOL(lookup_one_len);
2919 EXPORT_SYMBOL(page_follow_link_light);
2920 EXPORT_SYMBOL(page_put_link);
2921 EXPORT_SYMBOL(page_readlink);
2922 EXPORT_SYMBOL(__page_symlink);
2923 EXPORT_SYMBOL(page_symlink);
2924 EXPORT_SYMBOL(page_symlink_inode_operations);
2925 EXPORT_SYMBOL(path_lookup);
2926 EXPORT_SYMBOL(kern_path);
2927 EXPORT_SYMBOL(vfs_path_lookup);
2928 EXPORT_SYMBOL(inode_permission);
2929 EXPORT_SYMBOL(file_permission);
2930 EXPORT_SYMBOL(unlock_rename);
2931 EXPORT_SYMBOL(vfs_create);
2932 EXPORT_SYMBOL(vfs_follow_link);
2933 EXPORT_SYMBOL(vfs_link);
2934 EXPORT_SYMBOL(vfs_mkdir);
2935 EXPORT_SYMBOL(vfs_mknod);
2936 EXPORT_SYMBOL(generic_permission);
2937 EXPORT_SYMBOL(vfs_readlink);
2938 EXPORT_SYMBOL(vfs_rename);
2939 EXPORT_SYMBOL(vfs_rmdir);
2940 EXPORT_SYMBOL(vfs_symlink);
2941 EXPORT_SYMBOL(vfs_unlink);
2942 EXPORT_SYMBOL(dentry_unhash);
2943 EXPORT_SYMBOL(generic_readlink);