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 static int __link_path_walk(const char *name, struct nameidata *nd);
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static int do_getname(const char __user *filename, char *page)
123 unsigned long len = PATH_MAX;
125 if (!segment_eq(get_fs(), KERNEL_DS)) {
126 if ((unsigned long) filename >= TASK_SIZE)
128 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
129 len = TASK_SIZE - (unsigned long) filename;
132 retval = strncpy_from_user(page, filename, len);
136 return -ENAMETOOLONG;
142 char * getname(const char __user * filename)
146 result = ERR_PTR(-ENOMEM);
149 int retval = do_getname(filename, tmp);
154 result = ERR_PTR(retval);
157 audit_getname(result);
161 #ifdef CONFIG_AUDITSYSCALL
162 void putname(const char *name)
164 if (unlikely(!audit_dummy_context()))
169 EXPORT_SYMBOL(putname);
173 * This does basic POSIX ACL permission checking
175 static int acl_permission_check(struct inode *inode, int mask,
176 int (*check_acl)(struct inode *inode, int mask))
178 umode_t mode = inode->i_mode;
180 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
182 if (current_fsuid() == inode->i_uid)
185 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
191 if (in_group_p(inode->i_gid))
196 * If the DACs are ok we don't need any capability check.
198 if ((mask & ~mode) == 0)
204 * generic_permission - check for access rights on a Posix-like filesystem
205 * @inode: inode to check access rights for
206 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
207 * @check_acl: optional callback to check for Posix ACLs
209 * Used to check for read/write/execute permissions on a file.
210 * We use "fsuid" for this, letting us set arbitrary permissions
211 * for filesystem access without changing the "normal" uids which
212 * are used for other things..
214 int generic_permission(struct inode *inode, int mask,
215 int (*check_acl)(struct inode *inode, int mask))
220 * Do the basic POSIX ACL permission checks.
222 ret = acl_permission_check(inode, mask, check_acl);
227 * Read/write DACs are always overridable.
228 * Executable DACs are overridable if at least one exec bit is set.
230 if (!(mask & MAY_EXEC) || execute_ok(inode))
231 if (capable(CAP_DAC_OVERRIDE))
235 * Searching includes executable on directories, else just read.
237 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
238 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
239 if (capable(CAP_DAC_READ_SEARCH))
246 * inode_permission - check for access rights to a given inode
247 * @inode: inode to check permission on
248 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
250 * Used to check for read/write/execute permissions on an inode.
251 * We use "fsuid" for this, letting us set arbitrary permissions
252 * for filesystem access without changing the "normal" uids which
253 * are used for other things.
255 int inode_permission(struct inode *inode, int mask)
259 if (mask & MAY_WRITE) {
260 umode_t mode = inode->i_mode;
263 * Nobody gets write access to a read-only fs.
265 if (IS_RDONLY(inode) &&
266 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
270 * Nobody gets write access to an immutable file.
272 if (IS_IMMUTABLE(inode))
276 if (inode->i_op->permission)
277 retval = inode->i_op->permission(inode, mask);
279 retval = generic_permission(inode, mask, inode->i_op->check_acl);
284 retval = devcgroup_inode_permission(inode, mask);
288 return security_inode_permission(inode,
289 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
293 * file_permission - check for additional access rights to a given file
294 * @file: file to check access rights for
295 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
297 * Used to check for read/write/execute permissions on an already opened
301 * Do not use this function in new code. All access checks should
302 * be done using inode_permission().
304 int file_permission(struct file *file, int mask)
306 return inode_permission(file->f_path.dentry->d_inode, mask);
310 * get_write_access() gets write permission for a file.
311 * put_write_access() releases this write permission.
312 * This is used for regular files.
313 * We cannot support write (and maybe mmap read-write shared) accesses and
314 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
315 * can have the following values:
316 * 0: no writers, no VM_DENYWRITE mappings
317 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
318 * > 0: (i_writecount) users are writing to the file.
320 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
321 * except for the cases where we don't hold i_writecount yet. Then we need to
322 * use {get,deny}_write_access() - these functions check the sign and refuse
323 * to do the change if sign is wrong. Exclusion between them is provided by
324 * the inode->i_lock spinlock.
327 int get_write_access(struct inode * inode)
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) < 0) {
331 spin_unlock(&inode->i_lock);
334 atomic_inc(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
340 int deny_write_access(struct file * file)
342 struct inode *inode = file->f_path.dentry->d_inode;
344 spin_lock(&inode->i_lock);
345 if (atomic_read(&inode->i_writecount) > 0) {
346 spin_unlock(&inode->i_lock);
349 atomic_dec(&inode->i_writecount);
350 spin_unlock(&inode->i_lock);
356 * path_get - get a reference to a path
357 * @path: path to get the reference to
359 * Given a path increment the reference count to the dentry and the vfsmount.
361 void path_get(struct path *path)
366 EXPORT_SYMBOL(path_get);
369 * path_put - put a reference to a path
370 * @path: path to put the reference to
372 * Given a path decrement the reference count to the dentry and the vfsmount.
374 void path_put(struct path *path)
379 EXPORT_SYMBOL(path_put);
382 * release_open_intent - free up open intent resources
383 * @nd: pointer to nameidata
385 void release_open_intent(struct nameidata *nd)
387 if (nd->intent.open.file->f_path.dentry == NULL)
388 put_filp(nd->intent.open.file);
390 fput(nd->intent.open.file);
393 static inline struct dentry *
394 do_revalidate(struct dentry *dentry, struct nameidata *nd)
396 int status = dentry->d_op->d_revalidate(dentry, nd);
397 if (unlikely(status <= 0)) {
399 * The dentry failed validation.
400 * If d_revalidate returned 0 attempt to invalidate
401 * the dentry otherwise d_revalidate is asking us
402 * to return a fail status.
405 if (!d_invalidate(dentry)) {
411 dentry = ERR_PTR(status);
418 * Internal lookup() using the new generic dcache.
421 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
423 struct dentry * dentry = __d_lookup(parent, name);
425 /* lockess __d_lookup may fail due to concurrent d_move()
426 * in some unrelated directory, so try with d_lookup
429 dentry = d_lookup(parent, name);
431 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
432 dentry = do_revalidate(dentry, nd);
438 * Short-cut version of permission(), for calling by
439 * path_walk(), when dcache lock is held. Combines parts
440 * of permission() and generic_permission(), and tests ONLY for
441 * MAY_EXEC permission.
443 * If appropriate, check DAC only. If not appropriate, or
444 * short-cut DAC fails, then call permission() to do more
445 * complete permission check.
447 static int exec_permission_lite(struct inode *inode)
451 if (inode->i_op->permission) {
452 ret = inode->i_op->permission(inode, MAY_EXEC);
457 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
461 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
466 return security_inode_permission(inode, MAY_EXEC);
470 * This is called when everything else fails, and we actually have
471 * to go to the low-level filesystem to find out what we should do..
473 * We get the directory semaphore, and after getting that we also
474 * make sure that nobody added the entry to the dcache in the meantime..
477 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
479 struct dentry * result;
480 struct inode *dir = parent->d_inode;
482 mutex_lock(&dir->i_mutex);
484 * First re-do the cached lookup just in case it was created
485 * while we waited for the directory semaphore..
487 * FIXME! This could use version numbering or similar to
488 * avoid unnecessary cache lookups.
490 * The "dcache_lock" is purely to protect the RCU list walker
491 * from concurrent renames at this point (we mustn't get false
492 * negatives from the RCU list walk here, unlike the optimistic
495 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
497 result = d_lookup(parent, name);
499 struct dentry *dentry;
501 /* Don't create child dentry for a dead directory. */
502 result = ERR_PTR(-ENOENT);
506 dentry = d_alloc(parent, name);
507 result = ERR_PTR(-ENOMEM);
509 result = dir->i_op->lookup(dir, dentry, nd);
516 mutex_unlock(&dir->i_mutex);
521 * Uhhuh! Nasty case: the cache was re-populated while
522 * we waited on the semaphore. Need to revalidate.
524 mutex_unlock(&dir->i_mutex);
525 if (result->d_op && result->d_op->d_revalidate) {
526 result = do_revalidate(result, nd);
528 result = ERR_PTR(-ENOENT);
534 * Wrapper to retry pathname resolution whenever the underlying
535 * file system returns an ESTALE.
537 * Retry the whole path once, forcing real lookup requests
538 * instead of relying on the dcache.
540 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
542 struct path save = nd->path;
545 /* make sure the stuff we saved doesn't go away */
548 result = __link_path_walk(name, nd);
549 if (result == -ESTALE) {
550 /* nd->path had been dropped */
553 nd->flags |= LOOKUP_REVAL;
554 result = __link_path_walk(name, nd);
562 static __always_inline void set_root(struct nameidata *nd)
565 struct fs_struct *fs = current->fs;
566 read_lock(&fs->lock);
569 read_unlock(&fs->lock);
573 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
587 res = link_path_walk(link, nd);
588 if (nd->depth || res || nd->last_type!=LAST_NORM)
591 * If it is an iterative symlinks resolution in open_namei() we
592 * have to copy the last component. And all that crap because of
593 * bloody create() on broken symlinks. Furrfu...
596 if (unlikely(!name)) {
600 strcpy(name, nd->last.name);
601 nd->last.name = name;
605 return PTR_ERR(link);
608 static void path_put_conditional(struct path *path, struct nameidata *nd)
611 if (path->mnt != nd->path.mnt)
615 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
617 dput(nd->path.dentry);
618 if (nd->path.mnt != path->mnt)
619 mntput(nd->path.mnt);
620 nd->path.mnt = path->mnt;
621 nd->path.dentry = path->dentry;
624 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
628 struct dentry *dentry = path->dentry;
630 touch_atime(path->mnt, dentry);
631 nd_set_link(nd, NULL);
633 if (path->mnt != nd->path.mnt) {
634 path_to_nameidata(path, nd);
638 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
639 error = PTR_ERR(cookie);
640 if (!IS_ERR(cookie)) {
641 char *s = nd_get_link(nd);
644 error = __vfs_follow_link(nd, s);
645 if (dentry->d_inode->i_op->put_link)
646 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
654 * This limits recursive symlink follows to 8, while
655 * limiting consecutive symlinks to 40.
657 * Without that kind of total limit, nasty chains of consecutive
658 * symlinks can cause almost arbitrarily long lookups.
660 static inline int do_follow_link(struct path *path, struct nameidata *nd)
663 if (current->link_count >= MAX_NESTED_LINKS)
665 if (current->total_link_count >= 40)
667 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
669 err = security_inode_follow_link(path->dentry, nd);
672 current->link_count++;
673 current->total_link_count++;
675 err = __do_follow_link(path, nd);
676 current->link_count--;
680 path_put_conditional(path, nd);
685 int follow_up(struct path *path)
687 struct vfsmount *parent;
688 struct dentry *mountpoint;
689 spin_lock(&vfsmount_lock);
690 parent = path->mnt->mnt_parent;
691 if (parent == path->mnt) {
692 spin_unlock(&vfsmount_lock);
696 mountpoint = dget(path->mnt->mnt_mountpoint);
697 spin_unlock(&vfsmount_lock);
699 path->dentry = mountpoint;
705 /* no need for dcache_lock, as serialization is taken care in
708 static int __follow_mount(struct path *path)
711 while (d_mountpoint(path->dentry)) {
712 struct vfsmount *mounted = lookup_mnt(path);
719 path->dentry = dget(mounted->mnt_root);
725 static void follow_mount(struct path *path)
727 while (d_mountpoint(path->dentry)) {
728 struct vfsmount *mounted = lookup_mnt(path);
734 path->dentry = dget(mounted->mnt_root);
738 /* no need for dcache_lock, as serialization is taken care in
741 int follow_down(struct path *path)
743 struct vfsmount *mounted;
745 mounted = lookup_mnt(path);
750 path->dentry = dget(mounted->mnt_root);
756 static __always_inline void follow_dotdot(struct nameidata *nd)
761 struct vfsmount *parent;
762 struct dentry *old = nd->path.dentry;
764 if (nd->path.dentry == nd->root.dentry &&
765 nd->path.mnt == nd->root.mnt) {
768 spin_lock(&dcache_lock);
769 if (nd->path.dentry != nd->path.mnt->mnt_root) {
770 nd->path.dentry = dget(nd->path.dentry->d_parent);
771 spin_unlock(&dcache_lock);
775 spin_unlock(&dcache_lock);
776 spin_lock(&vfsmount_lock);
777 parent = nd->path.mnt->mnt_parent;
778 if (parent == nd->path.mnt) {
779 spin_unlock(&vfsmount_lock);
783 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
784 spin_unlock(&vfsmount_lock);
786 mntput(nd->path.mnt);
787 nd->path.mnt = parent;
789 follow_mount(&nd->path);
793 * It's more convoluted than I'd like it to be, but... it's still fairly
794 * small and for now I'd prefer to have fast path as straight as possible.
795 * It _is_ time-critical.
797 static int do_lookup(struct nameidata *nd, struct qstr *name,
800 struct vfsmount *mnt = nd->path.mnt;
801 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
805 if (dentry->d_op && dentry->d_op->d_revalidate)
806 goto need_revalidate;
809 path->dentry = dentry;
810 __follow_mount(path);
814 dentry = real_lookup(nd->path.dentry, name, nd);
820 dentry = do_revalidate(dentry, nd);
828 return PTR_ERR(dentry);
832 * This is a temporary kludge to deal with "automount" symlinks; proper
833 * solution is to trigger them on follow_mount(), so that do_lookup()
834 * would DTRT. To be killed before 2.6.34-final.
836 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
838 return inode && unlikely(inode->i_op->follow_link) &&
839 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
844 * This is the basic name resolution function, turning a pathname into
845 * the final dentry. We expect 'base' to be positive and a directory.
847 * Returns 0 and nd will have valid dentry and mnt on success.
848 * Returns error and drops reference to input namei data on failure.
850 static int __link_path_walk(const char *name, struct nameidata *nd)
855 unsigned int lookup_flags = nd->flags;
862 inode = nd->path.dentry->d_inode;
864 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
866 /* At this point we know we have a real path component. */
872 nd->flags |= LOOKUP_CONTINUE;
873 err = exec_permission_lite(inode);
878 c = *(const unsigned char *)name;
880 hash = init_name_hash();
883 hash = partial_name_hash(c, hash);
884 c = *(const unsigned char *)name;
885 } while (c && (c != '/'));
886 this.len = name - (const char *) this.name;
887 this.hash = end_name_hash(hash);
889 /* remove trailing slashes? */
892 while (*++name == '/');
894 goto last_with_slashes;
897 * "." and ".." are special - ".." especially so because it has
898 * to be able to know about the current root directory and
899 * parent relationships.
901 if (this.name[0] == '.') switch (this.len) {
905 if (this.name[1] != '.')
908 inode = nd->path.dentry->d_inode;
914 * See if the low-level filesystem might want
915 * to use its own hash..
917 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
918 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
923 /* This does the actual lookups.. */
924 err = do_lookup(nd, &this, &next);
929 inode = next.dentry->d_inode;
933 if (inode->i_op->follow_link) {
934 err = do_follow_link(&next, nd);
938 inode = nd->path.dentry->d_inode;
942 path_to_nameidata(&next, nd);
944 if (!inode->i_op->lookup)
947 /* here ends the main loop */
950 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
952 /* Clear LOOKUP_CONTINUE iff it was previously unset */
953 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
954 if (lookup_flags & LOOKUP_PARENT)
956 if (this.name[0] == '.') switch (this.len) {
960 if (this.name[1] != '.')
963 inode = nd->path.dentry->d_inode;
968 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
969 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
974 err = do_lookup(nd, &this, &next);
977 inode = next.dentry->d_inode;
978 if (follow_on_final(inode, lookup_flags)) {
979 err = do_follow_link(&next, nd);
982 inode = nd->path.dentry->d_inode;
984 path_to_nameidata(&next, nd);
988 if (lookup_flags & LOOKUP_DIRECTORY) {
990 if (!inode->i_op->lookup)
996 nd->last_type = LAST_NORM;
997 if (this.name[0] != '.')
1000 nd->last_type = LAST_DOT;
1001 else if (this.len == 2 && this.name[1] == '.')
1002 nd->last_type = LAST_DOTDOT;
1007 * We bypassed the ordinary revalidation routines.
1008 * We may need to check the cached dentry for staleness.
1010 if (nd->path.dentry && nd->path.dentry->d_sb &&
1011 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1013 /* Note: we do not d_invalidate() */
1014 if (!nd->path.dentry->d_op->d_revalidate(
1015 nd->path.dentry, nd))
1021 path_put_conditional(&next, nd);
1024 path_put(&nd->path);
1029 static int path_walk(const char *name, struct nameidata *nd)
1031 current->total_link_count = 0;
1032 return link_path_walk(name, nd);
1035 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1041 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1044 nd->root.mnt = NULL;
1048 nd->path = nd->root;
1049 path_get(&nd->root);
1050 } else if (dfd == AT_FDCWD) {
1051 struct fs_struct *fs = current->fs;
1052 read_lock(&fs->lock);
1055 read_unlock(&fs->lock);
1057 struct dentry *dentry;
1059 file = fget_light(dfd, &fput_needed);
1064 dentry = file->f_path.dentry;
1067 if (!S_ISDIR(dentry->d_inode->i_mode))
1070 retval = file_permission(file, MAY_EXEC);
1074 nd->path = file->f_path;
1075 path_get(&file->f_path);
1077 fput_light(file, fput_needed);
1082 fput_light(file, fput_needed);
1087 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1088 static int do_path_lookup(int dfd, const char *name,
1089 unsigned int flags, struct nameidata *nd)
1091 int retval = path_init(dfd, name, flags, nd);
1093 retval = path_walk(name, nd);
1094 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1095 nd->path.dentry->d_inode))
1096 audit_inode(name, nd->path.dentry);
1098 path_put(&nd->root);
1099 nd->root.mnt = NULL;
1104 int path_lookup(const char *name, unsigned int flags,
1105 struct nameidata *nd)
1107 return do_path_lookup(AT_FDCWD, name, flags, nd);
1110 int kern_path(const char *name, unsigned int flags, struct path *path)
1112 struct nameidata nd;
1113 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1120 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1121 * @dentry: pointer to dentry of the base directory
1122 * @mnt: pointer to vfs mount of the base directory
1123 * @name: pointer to file name
1124 * @flags: lookup flags
1125 * @nd: pointer to nameidata
1127 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1128 const char *name, unsigned int flags,
1129 struct nameidata *nd)
1133 /* same as do_path_lookup */
1134 nd->last_type = LAST_ROOT;
1138 nd->path.dentry = dentry;
1140 path_get(&nd->path);
1141 nd->root = nd->path;
1142 path_get(&nd->root);
1144 retval = path_walk(name, nd);
1145 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1146 nd->path.dentry->d_inode))
1147 audit_inode(name, nd->path.dentry);
1149 path_put(&nd->root);
1150 nd->root.mnt = NULL;
1156 * path_lookup_open - lookup a file path with open intent
1157 * @dfd: the directory to use as base, or AT_FDCWD
1158 * @name: pointer to file name
1159 * @lookup_flags: lookup intent flags
1160 * @nd: pointer to nameidata
1161 * @open_flags: open intent flags
1163 static int path_lookup_open(int dfd, const char *name,
1164 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1166 struct file *filp = get_empty_filp();
1171 nd->intent.open.file = filp;
1172 nd->intent.open.flags = open_flags;
1173 nd->intent.open.create_mode = 0;
1174 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1175 if (IS_ERR(nd->intent.open.file)) {
1177 err = PTR_ERR(nd->intent.open.file);
1178 path_put(&nd->path);
1180 } else if (err != 0)
1181 release_open_intent(nd);
1185 static struct dentry *__lookup_hash(struct qstr *name,
1186 struct dentry *base, struct nameidata *nd)
1188 struct dentry *dentry;
1189 struct inode *inode;
1192 inode = base->d_inode;
1195 * See if the low-level filesystem might want
1196 * to use its own hash..
1198 if (base->d_op && base->d_op->d_hash) {
1199 err = base->d_op->d_hash(base, name);
1200 dentry = ERR_PTR(err);
1205 dentry = cached_lookup(base, name, nd);
1209 /* Don't create child dentry for a dead directory. */
1210 dentry = ERR_PTR(-ENOENT);
1211 if (IS_DEADDIR(inode))
1214 new = d_alloc(base, name);
1215 dentry = ERR_PTR(-ENOMEM);
1218 dentry = inode->i_op->lookup(inode, new, nd);
1229 * Restricted form of lookup. Doesn't follow links, single-component only,
1230 * needs parent already locked. Doesn't follow mounts.
1233 static struct dentry *lookup_hash(struct nameidata *nd)
1237 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1239 return ERR_PTR(err);
1240 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1243 static int __lookup_one_len(const char *name, struct qstr *this,
1244 struct dentry *base, int len)
1254 hash = init_name_hash();
1256 c = *(const unsigned char *)name++;
1257 if (c == '/' || c == '\0')
1259 hash = partial_name_hash(c, hash);
1261 this->hash = end_name_hash(hash);
1266 * lookup_one_len - filesystem helper to lookup single pathname component
1267 * @name: pathname component to lookup
1268 * @base: base directory to lookup from
1269 * @len: maximum length @len should be interpreted to
1271 * Note that this routine is purely a helper for filesystem usage and should
1272 * not be called by generic code. Also note that by using this function the
1273 * nameidata argument is passed to the filesystem methods and a filesystem
1274 * using this helper needs to be prepared for that.
1276 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1281 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1283 err = __lookup_one_len(name, &this, base, len);
1285 return ERR_PTR(err);
1287 err = inode_permission(base->d_inode, MAY_EXEC);
1289 return ERR_PTR(err);
1290 return __lookup_hash(&this, base, NULL);
1294 * lookup_one_noperm - bad hack for sysfs
1295 * @name: pathname component to lookup
1296 * @base: base directory to lookup from
1298 * This is a variant of lookup_one_len that doesn't perform any permission
1299 * checks. It's a horrible hack to work around the braindead sysfs
1300 * architecture and should not be used anywhere else.
1302 * DON'T USE THIS FUNCTION EVER, thanks.
1304 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1309 err = __lookup_one_len(name, &this, base, strlen(name));
1311 return ERR_PTR(err);
1312 return __lookup_hash(&this, base, NULL);
1315 int user_path_at(int dfd, const char __user *name, unsigned flags,
1318 struct nameidata nd;
1319 char *tmp = getname(name);
1320 int err = PTR_ERR(tmp);
1323 BUG_ON(flags & LOOKUP_PARENT);
1325 err = do_path_lookup(dfd, tmp, flags, &nd);
1333 static int user_path_parent(int dfd, const char __user *path,
1334 struct nameidata *nd, char **name)
1336 char *s = getname(path);
1342 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1352 * It's inline, so penalty for filesystems that don't use sticky bit is
1355 static inline int check_sticky(struct inode *dir, struct inode *inode)
1357 uid_t fsuid = current_fsuid();
1359 if (!(dir->i_mode & S_ISVTX))
1361 if (inode->i_uid == fsuid)
1363 if (dir->i_uid == fsuid)
1365 return !capable(CAP_FOWNER);
1369 * Check whether we can remove a link victim from directory dir, check
1370 * whether the type of victim is right.
1371 * 1. We can't do it if dir is read-only (done in permission())
1372 * 2. We should have write and exec permissions on dir
1373 * 3. We can't remove anything from append-only dir
1374 * 4. We can't do anything with immutable dir (done in permission())
1375 * 5. If the sticky bit on dir is set we should either
1376 * a. be owner of dir, or
1377 * b. be owner of victim, or
1378 * c. have CAP_FOWNER capability
1379 * 6. If the victim is append-only or immutable we can't do antyhing with
1380 * links pointing to it.
1381 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1382 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1383 * 9. We can't remove a root or mountpoint.
1384 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1385 * nfs_async_unlink().
1387 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1391 if (!victim->d_inode)
1394 BUG_ON(victim->d_parent->d_inode != dir);
1395 audit_inode_child(victim->d_name.name, victim, dir);
1397 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1402 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1403 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1406 if (!S_ISDIR(victim->d_inode->i_mode))
1408 if (IS_ROOT(victim))
1410 } else if (S_ISDIR(victim->d_inode->i_mode))
1412 if (IS_DEADDIR(dir))
1414 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1419 /* Check whether we can create an object with dentry child in directory
1421 * 1. We can't do it if child already exists (open has special treatment for
1422 * this case, but since we are inlined it's OK)
1423 * 2. We can't do it if dir is read-only (done in permission())
1424 * 3. We should have write and exec permissions on dir
1425 * 4. We can't do it if dir is immutable (done in permission())
1427 static inline int may_create(struct inode *dir, struct dentry *child)
1431 if (IS_DEADDIR(dir))
1433 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1437 * O_DIRECTORY translates into forcing a directory lookup.
1439 static inline int lookup_flags(unsigned int f)
1441 unsigned long retval = LOOKUP_FOLLOW;
1444 retval &= ~LOOKUP_FOLLOW;
1446 if (f & O_DIRECTORY)
1447 retval |= LOOKUP_DIRECTORY;
1453 * p1 and p2 should be directories on the same fs.
1455 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1460 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1464 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1466 p = d_ancestor(p2, p1);
1468 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1469 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1473 p = d_ancestor(p1, p2);
1475 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1476 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1480 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1481 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1485 void unlock_rename(struct dentry *p1, struct dentry *p2)
1487 mutex_unlock(&p1->d_inode->i_mutex);
1489 mutex_unlock(&p2->d_inode->i_mutex);
1490 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1494 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1495 struct nameidata *nd)
1497 int error = may_create(dir, dentry);
1502 if (!dir->i_op->create)
1503 return -EACCES; /* shouldn't it be ENOSYS? */
1506 error = security_inode_create(dir, dentry, mode);
1510 error = dir->i_op->create(dir, dentry, mode, nd);
1512 fsnotify_create(dir, dentry);
1516 int may_open(struct path *path, int acc_mode, int flag)
1518 struct dentry *dentry = path->dentry;
1519 struct inode *inode = dentry->d_inode;
1525 switch (inode->i_mode & S_IFMT) {
1529 if (acc_mode & MAY_WRITE)
1534 if (path->mnt->mnt_flags & MNT_NODEV)
1543 error = inode_permission(inode, acc_mode);
1547 error = ima_path_check(path, acc_mode ?
1548 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1549 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1555 * An append-only file must be opened in append mode for writing.
1557 if (IS_APPEND(inode)) {
1559 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1565 /* O_NOATIME can only be set by the owner or superuser */
1566 if (flag & O_NOATIME)
1567 if (!is_owner_or_cap(inode)) {
1573 * Ensure there are no outstanding leases on the file.
1575 error = break_lease(inode, flag);
1579 if (flag & O_TRUNC) {
1580 error = get_write_access(inode);
1585 * Refuse to truncate files with mandatory locks held on them.
1587 error = locks_verify_locked(inode);
1589 error = security_path_truncate(path, 0,
1590 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1594 error = do_truncate(dentry, 0,
1595 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1598 put_write_access(inode);
1602 if (flag & FMODE_WRITE)
1607 ima_counts_put(path, acc_mode ?
1608 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1609 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1614 * Be careful about ever adding any more callers of this
1615 * function. Its flags must be in the namei format, not
1616 * what get passed to sys_open().
1618 static int __open_namei_create(struct nameidata *nd, struct path *path,
1622 struct dentry *dir = nd->path.dentry;
1624 if (!IS_POSIXACL(dir->d_inode))
1625 mode &= ~current_umask();
1626 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1629 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1631 mutex_unlock(&dir->d_inode->i_mutex);
1632 dput(nd->path.dentry);
1633 nd->path.dentry = path->dentry;
1636 /* Don't check for write permission, don't truncate */
1637 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1641 * Note that while the flag value (low two bits) for sys_open means:
1646 * it is changed into
1647 * 00 - no permissions needed
1648 * 01 - read-permission
1649 * 10 - write-permission
1651 * for the internal routines (ie open_namei()/follow_link() etc)
1652 * This is more logical, and also allows the 00 "no perm needed"
1653 * to be used for symlinks (where the permissions are checked
1657 static inline int open_to_namei_flags(int flag)
1659 if ((flag+1) & O_ACCMODE)
1664 static int open_will_write_to_fs(int flag, struct inode *inode)
1667 * We'll never write to the fs underlying
1670 if (special_file(inode->i_mode))
1672 return (flag & O_TRUNC);
1676 * Note that the low bits of the passed in "open_flag"
1677 * are not the same as in the local variable "flag". See
1678 * open_to_namei_flags() for more details.
1680 struct file *do_filp_open(int dfd, const char *pathname,
1681 int open_flag, int mode, int acc_mode)
1684 struct nameidata nd;
1690 int flag = open_to_namei_flags(open_flag);
1693 acc_mode = MAY_OPEN | ACC_MODE(flag);
1695 /* O_TRUNC implies we need access checks for write permissions */
1697 acc_mode |= MAY_WRITE;
1699 /* Allow the LSM permission hook to distinguish append
1700 access from general write access. */
1701 if (flag & O_APPEND)
1702 acc_mode |= MAY_APPEND;
1705 * The simplest case - just a plain lookup.
1707 if (!(flag & O_CREAT)) {
1708 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1711 return ERR_PTR(error);
1716 * Create - we need to know the parent.
1718 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1720 return ERR_PTR(error);
1721 error = path_walk(pathname, &nd);
1725 return ERR_PTR(error);
1727 if (unlikely(!audit_dummy_context()))
1728 audit_inode(pathname, nd.path.dentry);
1731 * We have the parent and last component. First of all, check
1732 * that we are not asked to creat(2) an obvious directory - that
1736 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1740 filp = get_empty_filp();
1743 nd.intent.open.file = filp;
1744 nd.intent.open.flags = flag;
1745 nd.intent.open.create_mode = mode;
1746 dir = nd.path.dentry;
1747 nd.flags &= ~LOOKUP_PARENT;
1748 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1750 nd.flags |= LOOKUP_EXCL;
1751 mutex_lock(&dir->d_inode->i_mutex);
1752 path.dentry = lookup_hash(&nd);
1753 path.mnt = nd.path.mnt;
1756 error = PTR_ERR(path.dentry);
1757 if (IS_ERR(path.dentry)) {
1758 mutex_unlock(&dir->d_inode->i_mutex);
1762 if (IS_ERR(nd.intent.open.file)) {
1763 error = PTR_ERR(nd.intent.open.file);
1764 goto exit_mutex_unlock;
1767 /* Negative dentry, just create the file */
1768 if (!path.dentry->d_inode) {
1770 * This write is needed to ensure that a
1771 * ro->rw transition does not occur between
1772 * the time when the file is created and when
1773 * a permanent write count is taken through
1774 * the 'struct file' in nameidata_to_filp().
1776 error = mnt_want_write(nd.path.mnt);
1778 goto exit_mutex_unlock;
1779 error = __open_namei_create(&nd, &path, flag, mode);
1781 mnt_drop_write(nd.path.mnt);
1784 filp = nameidata_to_filp(&nd, open_flag);
1786 ima_counts_put(&nd.path,
1787 acc_mode & (MAY_READ | MAY_WRITE |
1789 mnt_drop_write(nd.path.mnt);
1796 * It already exists.
1798 mutex_unlock(&dir->d_inode->i_mutex);
1799 audit_inode(pathname, path.dentry);
1805 if (__follow_mount(&path)) {
1807 if (flag & O_NOFOLLOW)
1812 if (!path.dentry->d_inode)
1814 if (path.dentry->d_inode->i_op->follow_link)
1817 path_to_nameidata(&path, &nd);
1819 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1824 * 1. may_open() truncates a file
1825 * 2. a rw->ro mount transition occurs
1826 * 3. nameidata_to_filp() fails due to
1828 * That would be inconsistent, and should
1829 * be avoided. Taking this mnt write here
1830 * ensures that (2) can not occur.
1832 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1834 error = mnt_want_write(nd.path.mnt);
1838 error = may_open(&nd.path, acc_mode, flag);
1841 mnt_drop_write(nd.path.mnt);
1844 filp = nameidata_to_filp(&nd, open_flag);
1846 ima_counts_put(&nd.path,
1847 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1849 * It is now safe to drop the mnt write
1850 * because the filp has had a write taken
1854 mnt_drop_write(nd.path.mnt);
1860 mutex_unlock(&dir->d_inode->i_mutex);
1862 path_put_conditional(&path, &nd);
1864 if (!IS_ERR(nd.intent.open.file))
1865 release_open_intent(&nd);
1870 return ERR_PTR(error);
1874 if (flag & O_NOFOLLOW)
1877 * This is subtle. Instead of calling do_follow_link() we do the
1878 * thing by hands. The reason is that this way we have zero link_count
1879 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1880 * After that we have the parent and last component, i.e.
1881 * we are in the same situation as after the first path_walk().
1882 * Well, almost - if the last component is normal we get its copy
1883 * stored in nd->last.name and we will have to putname() it when we
1884 * are done. Procfs-like symlinks just set LAST_BIND.
1886 nd.flags |= LOOKUP_PARENT;
1887 error = security_inode_follow_link(path.dentry, &nd);
1890 error = __do_follow_link(&path, &nd);
1892 /* Does someone understand code flow here? Or it is only
1893 * me so stupid? Anathema to whoever designed this non-sense
1894 * with "intent.open".
1896 release_open_intent(&nd);
1899 return ERR_PTR(error);
1901 nd.flags &= ~LOOKUP_PARENT;
1902 if (nd.last_type == LAST_BIND)
1905 if (nd.last_type != LAST_NORM)
1907 if (nd.last.name[nd.last.len]) {
1908 __putname(nd.last.name);
1913 __putname(nd.last.name);
1916 dir = nd.path.dentry;
1917 mutex_lock(&dir->d_inode->i_mutex);
1918 path.dentry = lookup_hash(&nd);
1919 path.mnt = nd.path.mnt;
1920 __putname(nd.last.name);
1925 * filp_open - open file and return file pointer
1927 * @filename: path to open
1928 * @flags: open flags as per the open(2) second argument
1929 * @mode: mode for the new file if O_CREAT is set, else ignored
1931 * This is the helper to open a file from kernelspace if you really
1932 * have to. But in generally you should not do this, so please move
1933 * along, nothing to see here..
1935 struct file *filp_open(const char *filename, int flags, int mode)
1937 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1939 EXPORT_SYMBOL(filp_open);
1942 * lookup_create - lookup a dentry, creating it if it doesn't exist
1943 * @nd: nameidata info
1944 * @is_dir: directory flag
1946 * Simple function to lookup and return a dentry and create it
1947 * if it doesn't exist. Is SMP-safe.
1949 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1951 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1953 struct dentry *dentry = ERR_PTR(-EEXIST);
1955 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1957 * Yucky last component or no last component at all?
1958 * (foo/., foo/.., /////)
1960 if (nd->last_type != LAST_NORM)
1962 nd->flags &= ~LOOKUP_PARENT;
1963 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1964 nd->intent.open.flags = O_EXCL;
1967 * Do the final lookup.
1969 dentry = lookup_hash(nd);
1973 if (dentry->d_inode)
1976 * Special case - lookup gave negative, but... we had foo/bar/
1977 * From the vfs_mknod() POV we just have a negative dentry -
1978 * all is fine. Let's be bastards - you had / on the end, you've
1979 * been asking for (non-existent) directory. -ENOENT for you.
1981 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1983 dentry = ERR_PTR(-ENOENT);
1988 dentry = ERR_PTR(-EEXIST);
1992 EXPORT_SYMBOL_GPL(lookup_create);
1994 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1996 int error = may_create(dir, dentry);
2001 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2004 if (!dir->i_op->mknod)
2007 error = devcgroup_inode_mknod(mode, dev);
2011 error = security_inode_mknod(dir, dentry, mode, dev);
2016 error = dir->i_op->mknod(dir, dentry, mode, dev);
2018 fsnotify_create(dir, dentry);
2022 static int may_mknod(mode_t mode)
2024 switch (mode & S_IFMT) {
2030 case 0: /* zero mode translates to S_IFREG */
2039 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2044 struct dentry *dentry;
2045 struct nameidata nd;
2050 error = user_path_parent(dfd, filename, &nd, &tmp);
2054 dentry = lookup_create(&nd, 0);
2055 if (IS_ERR(dentry)) {
2056 error = PTR_ERR(dentry);
2059 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2060 mode &= ~current_umask();
2061 error = may_mknod(mode);
2064 error = mnt_want_write(nd.path.mnt);
2067 error = security_path_mknod(&nd.path, dentry, mode, dev);
2069 goto out_drop_write;
2070 switch (mode & S_IFMT) {
2071 case 0: case S_IFREG:
2072 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2074 case S_IFCHR: case S_IFBLK:
2075 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2076 new_decode_dev(dev));
2078 case S_IFIFO: case S_IFSOCK:
2079 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2083 mnt_drop_write(nd.path.mnt);
2087 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2094 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2096 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2099 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2101 int error = may_create(dir, dentry);
2106 if (!dir->i_op->mkdir)
2109 mode &= (S_IRWXUGO|S_ISVTX);
2110 error = security_inode_mkdir(dir, dentry, mode);
2115 error = dir->i_op->mkdir(dir, dentry, mode);
2117 fsnotify_mkdir(dir, dentry);
2121 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2125 struct dentry *dentry;
2126 struct nameidata nd;
2128 error = user_path_parent(dfd, pathname, &nd, &tmp);
2132 dentry = lookup_create(&nd, 1);
2133 error = PTR_ERR(dentry);
2137 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2138 mode &= ~current_umask();
2139 error = mnt_want_write(nd.path.mnt);
2142 error = security_path_mkdir(&nd.path, dentry, mode);
2144 goto out_drop_write;
2145 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2147 mnt_drop_write(nd.path.mnt);
2151 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2158 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2160 return sys_mkdirat(AT_FDCWD, pathname, mode);
2164 * We try to drop the dentry early: we should have
2165 * a usage count of 2 if we're the only user of this
2166 * dentry, and if that is true (possibly after pruning
2167 * the dcache), then we drop the dentry now.
2169 * A low-level filesystem can, if it choses, legally
2172 * if (!d_unhashed(dentry))
2175 * if it cannot handle the case of removing a directory
2176 * that is still in use by something else..
2178 void dentry_unhash(struct dentry *dentry)
2181 shrink_dcache_parent(dentry);
2182 spin_lock(&dcache_lock);
2183 spin_lock(&dentry->d_lock);
2184 if (atomic_read(&dentry->d_count) == 2)
2186 spin_unlock(&dentry->d_lock);
2187 spin_unlock(&dcache_lock);
2190 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2192 int error = may_delete(dir, dentry, 1);
2197 if (!dir->i_op->rmdir)
2202 mutex_lock(&dentry->d_inode->i_mutex);
2203 dentry_unhash(dentry);
2204 if (d_mountpoint(dentry))
2207 error = security_inode_rmdir(dir, dentry);
2209 error = dir->i_op->rmdir(dir, dentry);
2211 dentry->d_inode->i_flags |= S_DEAD;
2214 mutex_unlock(&dentry->d_inode->i_mutex);
2223 static long do_rmdir(int dfd, const char __user *pathname)
2227 struct dentry *dentry;
2228 struct nameidata nd;
2230 error = user_path_parent(dfd, pathname, &nd, &name);
2234 switch(nd.last_type) {
2246 nd.flags &= ~LOOKUP_PARENT;
2248 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2249 dentry = lookup_hash(&nd);
2250 error = PTR_ERR(dentry);
2253 error = mnt_want_write(nd.path.mnt);
2256 error = security_path_rmdir(&nd.path, dentry);
2259 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2261 mnt_drop_write(nd.path.mnt);
2265 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2272 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2274 return do_rmdir(AT_FDCWD, pathname);
2277 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2279 int error = may_delete(dir, dentry, 0);
2284 if (!dir->i_op->unlink)
2289 mutex_lock(&dentry->d_inode->i_mutex);
2290 if (d_mountpoint(dentry))
2293 error = security_inode_unlink(dir, dentry);
2295 error = dir->i_op->unlink(dir, dentry);
2297 mutex_unlock(&dentry->d_inode->i_mutex);
2299 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2300 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2301 fsnotify_link_count(dentry->d_inode);
2309 * Make sure that the actual truncation of the file will occur outside its
2310 * directory's i_mutex. Truncate can take a long time if there is a lot of
2311 * writeout happening, and we don't want to prevent access to the directory
2312 * while waiting on the I/O.
2314 static long do_unlinkat(int dfd, const char __user *pathname)
2318 struct dentry *dentry;
2319 struct nameidata nd;
2320 struct inode *inode = NULL;
2322 error = user_path_parent(dfd, pathname, &nd, &name);
2327 if (nd.last_type != LAST_NORM)
2330 nd.flags &= ~LOOKUP_PARENT;
2332 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2333 dentry = lookup_hash(&nd);
2334 error = PTR_ERR(dentry);
2335 if (!IS_ERR(dentry)) {
2336 /* Why not before? Because we want correct error value */
2337 if (nd.last.name[nd.last.len])
2339 inode = dentry->d_inode;
2341 atomic_inc(&inode->i_count);
2342 error = mnt_want_write(nd.path.mnt);
2345 error = security_path_unlink(&nd.path, dentry);
2348 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2350 mnt_drop_write(nd.path.mnt);
2354 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2356 iput(inode); /* truncate the inode here */
2363 error = !dentry->d_inode ? -ENOENT :
2364 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2368 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2370 if ((flag & ~AT_REMOVEDIR) != 0)
2373 if (flag & AT_REMOVEDIR)
2374 return do_rmdir(dfd, pathname);
2376 return do_unlinkat(dfd, pathname);
2379 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2381 return do_unlinkat(AT_FDCWD, pathname);
2384 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2386 int error = may_create(dir, dentry);
2391 if (!dir->i_op->symlink)
2394 error = security_inode_symlink(dir, dentry, oldname);
2399 error = dir->i_op->symlink(dir, dentry, oldname);
2401 fsnotify_create(dir, dentry);
2405 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2406 int, newdfd, const char __user *, newname)
2411 struct dentry *dentry;
2412 struct nameidata nd;
2414 from = getname(oldname);
2416 return PTR_ERR(from);
2418 error = user_path_parent(newdfd, newname, &nd, &to);
2422 dentry = lookup_create(&nd, 0);
2423 error = PTR_ERR(dentry);
2427 error = mnt_want_write(nd.path.mnt);
2430 error = security_path_symlink(&nd.path, dentry, from);
2432 goto out_drop_write;
2433 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2435 mnt_drop_write(nd.path.mnt);
2439 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2447 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2449 return sys_symlinkat(oldname, AT_FDCWD, newname);
2452 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2454 struct inode *inode = old_dentry->d_inode;
2460 error = may_create(dir, new_dentry);
2464 if (dir->i_sb != inode->i_sb)
2468 * A link to an append-only or immutable file cannot be created.
2470 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2472 if (!dir->i_op->link)
2474 if (S_ISDIR(inode->i_mode))
2477 error = security_inode_link(old_dentry, dir, new_dentry);
2481 mutex_lock(&inode->i_mutex);
2483 error = dir->i_op->link(old_dentry, dir, new_dentry);
2484 mutex_unlock(&inode->i_mutex);
2486 fsnotify_link(dir, inode, new_dentry);
2491 * Hardlinks are often used in delicate situations. We avoid
2492 * security-related surprises by not following symlinks on the
2495 * We don't follow them on the oldname either to be compatible
2496 * with linux 2.0, and to avoid hard-linking to directories
2497 * and other special files. --ADM
2499 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2500 int, newdfd, const char __user *, newname, int, flags)
2502 struct dentry *new_dentry;
2503 struct nameidata nd;
2504 struct path old_path;
2508 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2511 error = user_path_at(olddfd, oldname,
2512 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2517 error = user_path_parent(newdfd, newname, &nd, &to);
2521 if (old_path.mnt != nd.path.mnt)
2523 new_dentry = lookup_create(&nd, 0);
2524 error = PTR_ERR(new_dentry);
2525 if (IS_ERR(new_dentry))
2527 error = mnt_want_write(nd.path.mnt);
2530 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2532 goto out_drop_write;
2533 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2535 mnt_drop_write(nd.path.mnt);
2539 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2544 path_put(&old_path);
2549 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2551 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2555 * The worst of all namespace operations - renaming directory. "Perverted"
2556 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2558 * a) we can get into loop creation. Check is done in is_subdir().
2559 * b) race potential - two innocent renames can create a loop together.
2560 * That's where 4.4 screws up. Current fix: serialization on
2561 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2563 * c) we have to lock _three_ objects - parents and victim (if it exists).
2564 * And that - after we got ->i_mutex on parents (until then we don't know
2565 * whether the target exists). Solution: try to be smart with locking
2566 * order for inodes. We rely on the fact that tree topology may change
2567 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2568 * move will be locked. Thus we can rank directories by the tree
2569 * (ancestors first) and rank all non-directories after them.
2570 * That works since everybody except rename does "lock parent, lookup,
2571 * lock child" and rename is under ->s_vfs_rename_mutex.
2572 * HOWEVER, it relies on the assumption that any object with ->lookup()
2573 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2574 * we'd better make sure that there's no link(2) for them.
2575 * d) some filesystems don't support opened-but-unlinked directories,
2576 * either because of layout or because they are not ready to deal with
2577 * all cases correctly. The latter will be fixed (taking this sort of
2578 * stuff into VFS), but the former is not going away. Solution: the same
2579 * trick as in rmdir().
2580 * e) conversion from fhandle to dentry may come in the wrong moment - when
2581 * we are removing the target. Solution: we will have to grab ->i_mutex
2582 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2583 * ->i_mutex on parents, which works but leads to some truely excessive
2586 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2587 struct inode *new_dir, struct dentry *new_dentry)
2590 struct inode *target;
2593 * If we are going to change the parent - check write permissions,
2594 * we'll need to flip '..'.
2596 if (new_dir != old_dir) {
2597 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2602 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2606 target = new_dentry->d_inode;
2608 mutex_lock(&target->i_mutex);
2609 dentry_unhash(new_dentry);
2611 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2614 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2617 target->i_flags |= S_DEAD;
2618 mutex_unlock(&target->i_mutex);
2619 if (d_unhashed(new_dentry))
2620 d_rehash(new_dentry);
2624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2625 d_move(old_dentry,new_dentry);
2629 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2630 struct inode *new_dir, struct dentry *new_dentry)
2632 struct inode *target;
2635 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2640 target = new_dentry->d_inode;
2642 mutex_lock(&target->i_mutex);
2643 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2646 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2648 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2649 d_move(old_dentry, new_dentry);
2652 mutex_unlock(&target->i_mutex);
2657 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2658 struct inode *new_dir, struct dentry *new_dentry)
2661 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2662 const char *old_name;
2664 if (old_dentry->d_inode == new_dentry->d_inode)
2667 error = may_delete(old_dir, old_dentry, is_dir);
2671 if (!new_dentry->d_inode)
2672 error = may_create(new_dir, new_dentry);
2674 error = may_delete(new_dir, new_dentry, is_dir);
2678 if (!old_dir->i_op->rename)
2681 vfs_dq_init(old_dir);
2682 vfs_dq_init(new_dir);
2684 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2687 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2689 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2691 const char *new_name = old_dentry->d_name.name;
2692 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2693 new_dentry->d_inode, old_dentry);
2695 fsnotify_oldname_free(old_name);
2700 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2701 int, newdfd, const char __user *, newname)
2703 struct dentry *old_dir, *new_dir;
2704 struct dentry *old_dentry, *new_dentry;
2705 struct dentry *trap;
2706 struct nameidata oldnd, newnd;
2711 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2715 error = user_path_parent(newdfd, newname, &newnd, &to);
2720 if (oldnd.path.mnt != newnd.path.mnt)
2723 old_dir = oldnd.path.dentry;
2725 if (oldnd.last_type != LAST_NORM)
2728 new_dir = newnd.path.dentry;
2729 if (newnd.last_type != LAST_NORM)
2732 oldnd.flags &= ~LOOKUP_PARENT;
2733 newnd.flags &= ~LOOKUP_PARENT;
2734 newnd.flags |= LOOKUP_RENAME_TARGET;
2736 trap = lock_rename(new_dir, old_dir);
2738 old_dentry = lookup_hash(&oldnd);
2739 error = PTR_ERR(old_dentry);
2740 if (IS_ERR(old_dentry))
2742 /* source must exist */
2744 if (!old_dentry->d_inode)
2746 /* unless the source is a directory trailing slashes give -ENOTDIR */
2747 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2749 if (oldnd.last.name[oldnd.last.len])
2751 if (newnd.last.name[newnd.last.len])
2754 /* source should not be ancestor of target */
2756 if (old_dentry == trap)
2758 new_dentry = lookup_hash(&newnd);
2759 error = PTR_ERR(new_dentry);
2760 if (IS_ERR(new_dentry))
2762 /* target should not be an ancestor of source */
2764 if (new_dentry == trap)
2767 error = mnt_want_write(oldnd.path.mnt);
2770 error = security_path_rename(&oldnd.path, old_dentry,
2771 &newnd.path, new_dentry);
2774 error = vfs_rename(old_dir->d_inode, old_dentry,
2775 new_dir->d_inode, new_dentry);
2777 mnt_drop_write(oldnd.path.mnt);
2783 unlock_rename(new_dir, old_dir);
2785 path_put(&newnd.path);
2788 path_put(&oldnd.path);
2794 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2796 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2799 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2803 len = PTR_ERR(link);
2808 if (len > (unsigned) buflen)
2810 if (copy_to_user(buffer, link, len))
2817 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2818 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2819 * using) it for any given inode is up to filesystem.
2821 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2823 struct nameidata nd;
2828 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2830 return PTR_ERR(cookie);
2832 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2833 if (dentry->d_inode->i_op->put_link)
2834 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2838 int vfs_follow_link(struct nameidata *nd, const char *link)
2840 return __vfs_follow_link(nd, link);
2843 /* get the link contents into pagecache */
2844 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2848 struct address_space *mapping = dentry->d_inode->i_mapping;
2849 page = read_mapping_page(mapping, 0, NULL);
2854 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2858 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2860 struct page *page = NULL;
2861 char *s = page_getlink(dentry, &page);
2862 int res = vfs_readlink(dentry,buffer,buflen,s);
2865 page_cache_release(page);
2870 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2872 struct page *page = NULL;
2873 nd_set_link(nd, page_getlink(dentry, &page));
2877 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2879 struct page *page = cookie;
2883 page_cache_release(page);
2888 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2890 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2892 struct address_space *mapping = inode->i_mapping;
2897 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2899 flags |= AOP_FLAG_NOFS;
2902 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2903 flags, &page, &fsdata);
2907 kaddr = kmap_atomic(page, KM_USER0);
2908 memcpy(kaddr, symname, len-1);
2909 kunmap_atomic(kaddr, KM_USER0);
2911 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2918 mark_inode_dirty(inode);
2924 int page_symlink(struct inode *inode, const char *symname, int len)
2926 return __page_symlink(inode, symname, len,
2927 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2930 const struct inode_operations page_symlink_inode_operations = {
2931 .readlink = generic_readlink,
2932 .follow_link = page_follow_link_light,
2933 .put_link = page_put_link,
2936 EXPORT_SYMBOL(user_path_at);
2937 EXPORT_SYMBOL(follow_down);
2938 EXPORT_SYMBOL(follow_up);
2939 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2940 EXPORT_SYMBOL(getname);
2941 EXPORT_SYMBOL(lock_rename);
2942 EXPORT_SYMBOL(lookup_one_len);
2943 EXPORT_SYMBOL(page_follow_link_light);
2944 EXPORT_SYMBOL(page_put_link);
2945 EXPORT_SYMBOL(page_readlink);
2946 EXPORT_SYMBOL(__page_symlink);
2947 EXPORT_SYMBOL(page_symlink);
2948 EXPORT_SYMBOL(page_symlink_inode_operations);
2949 EXPORT_SYMBOL(path_lookup);
2950 EXPORT_SYMBOL(kern_path);
2951 EXPORT_SYMBOL(vfs_path_lookup);
2952 EXPORT_SYMBOL(inode_permission);
2953 EXPORT_SYMBOL(file_permission);
2954 EXPORT_SYMBOL(unlock_rename);
2955 EXPORT_SYMBOL(vfs_create);
2956 EXPORT_SYMBOL(vfs_follow_link);
2957 EXPORT_SYMBOL(vfs_link);
2958 EXPORT_SYMBOL(vfs_mkdir);
2959 EXPORT_SYMBOL(vfs_mknod);
2960 EXPORT_SYMBOL(generic_permission);
2961 EXPORT_SYMBOL(vfs_readlink);
2962 EXPORT_SYMBOL(vfs_rename);
2963 EXPORT_SYMBOL(vfs_rmdir);
2964 EXPORT_SYMBOL(vfs_symlink);
2965 EXPORT_SYMBOL(vfs_unlink);
2966 EXPORT_SYMBOL(dentry_unhash);
2967 EXPORT_SYMBOL(generic_readlink);