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);
833 * This is the basic name resolution function, turning a pathname into
834 * the final dentry. We expect 'base' to be positive and a directory.
836 * Returns 0 and nd will have valid dentry and mnt on success.
837 * Returns error and drops reference to input namei data on failure.
839 static int __link_path_walk(const char *name, struct nameidata *nd)
844 unsigned int lookup_flags = nd->flags;
851 inode = nd->path.dentry->d_inode;
853 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
855 /* At this point we know we have a real path component. */
861 nd->flags |= LOOKUP_CONTINUE;
862 err = exec_permission_lite(inode);
867 c = *(const unsigned char *)name;
869 hash = init_name_hash();
872 hash = partial_name_hash(c, hash);
873 c = *(const unsigned char *)name;
874 } while (c && (c != '/'));
875 this.len = name - (const char *) this.name;
876 this.hash = end_name_hash(hash);
878 /* remove trailing slashes? */
881 while (*++name == '/');
883 goto last_with_slashes;
886 * "." and ".." are special - ".." especially so because it has
887 * to be able to know about the current root directory and
888 * parent relationships.
890 if (this.name[0] == '.') switch (this.len) {
894 if (this.name[1] != '.')
897 inode = nd->path.dentry->d_inode;
903 * See if the low-level filesystem might want
904 * to use its own hash..
906 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
907 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
912 /* This does the actual lookups.. */
913 err = do_lookup(nd, &this, &next);
918 inode = next.dentry->d_inode;
922 if (inode->i_op->follow_link) {
923 err = do_follow_link(&next, nd);
927 inode = nd->path.dentry->d_inode;
931 path_to_nameidata(&next, nd);
933 if (!inode->i_op->lookup)
936 /* here ends the main loop */
939 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
941 /* Clear LOOKUP_CONTINUE iff it was previously unset */
942 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
943 if (lookup_flags & LOOKUP_PARENT)
945 if (this.name[0] == '.') switch (this.len) {
949 if (this.name[1] != '.')
952 inode = nd->path.dentry->d_inode;
957 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
958 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
963 err = do_lookup(nd, &this, &next);
966 inode = next.dentry->d_inode;
967 if ((lookup_flags & LOOKUP_FOLLOW)
968 && inode && inode->i_op->follow_link) {
969 err = do_follow_link(&next, nd);
972 inode = nd->path.dentry->d_inode;
974 path_to_nameidata(&next, nd);
978 if (lookup_flags & LOOKUP_DIRECTORY) {
980 if (!inode->i_op->lookup)
986 nd->last_type = LAST_NORM;
987 if (this.name[0] != '.')
990 nd->last_type = LAST_DOT;
991 else if (this.len == 2 && this.name[1] == '.')
992 nd->last_type = LAST_DOTDOT;
997 * We bypassed the ordinary revalidation routines.
998 * We may need to check the cached dentry for staleness.
1000 if (nd->path.dentry && nd->path.dentry->d_sb &&
1001 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1003 /* Note: we do not d_invalidate() */
1004 if (!nd->path.dentry->d_op->d_revalidate(
1005 nd->path.dentry, nd))
1011 path_put_conditional(&next, nd);
1014 path_put(&nd->path);
1019 static int path_walk(const char *name, struct nameidata *nd)
1021 current->total_link_count = 0;
1022 return link_path_walk(name, nd);
1025 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1031 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1034 nd->root.mnt = NULL;
1038 nd->path = nd->root;
1039 path_get(&nd->root);
1040 } else if (dfd == AT_FDCWD) {
1041 struct fs_struct *fs = current->fs;
1042 read_lock(&fs->lock);
1045 read_unlock(&fs->lock);
1047 struct dentry *dentry;
1049 file = fget_light(dfd, &fput_needed);
1054 dentry = file->f_path.dentry;
1057 if (!S_ISDIR(dentry->d_inode->i_mode))
1060 retval = file_permission(file, MAY_EXEC);
1064 nd->path = file->f_path;
1065 path_get(&file->f_path);
1067 fput_light(file, fput_needed);
1072 fput_light(file, fput_needed);
1077 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1078 static int do_path_lookup(int dfd, const char *name,
1079 unsigned int flags, struct nameidata *nd)
1081 int retval = path_init(dfd, name, flags, nd);
1083 retval = path_walk(name, nd);
1084 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1085 nd->path.dentry->d_inode))
1086 audit_inode(name, nd->path.dentry);
1088 path_put(&nd->root);
1089 nd->root.mnt = NULL;
1094 int path_lookup(const char *name, unsigned int flags,
1095 struct nameidata *nd)
1097 return do_path_lookup(AT_FDCWD, name, flags, nd);
1100 int kern_path(const char *name, unsigned int flags, struct path *path)
1102 struct nameidata nd;
1103 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1110 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1111 * @dentry: pointer to dentry of the base directory
1112 * @mnt: pointer to vfs mount of the base directory
1113 * @name: pointer to file name
1114 * @flags: lookup flags
1115 * @nd: pointer to nameidata
1117 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1118 const char *name, unsigned int flags,
1119 struct nameidata *nd)
1123 /* same as do_path_lookup */
1124 nd->last_type = LAST_ROOT;
1128 nd->path.dentry = dentry;
1130 path_get(&nd->path);
1131 nd->root = nd->path;
1132 path_get(&nd->root);
1134 retval = path_walk(name, nd);
1135 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1136 nd->path.dentry->d_inode))
1137 audit_inode(name, nd->path.dentry);
1139 path_put(&nd->root);
1140 nd->root.mnt = NULL;
1146 * path_lookup_open - lookup a file path with open intent
1147 * @dfd: the directory to use as base, or AT_FDCWD
1148 * @name: pointer to file name
1149 * @lookup_flags: lookup intent flags
1150 * @nd: pointer to nameidata
1151 * @open_flags: open intent flags
1153 static int path_lookup_open(int dfd, const char *name,
1154 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1156 struct file *filp = get_empty_filp();
1161 nd->intent.open.file = filp;
1162 nd->intent.open.flags = open_flags;
1163 nd->intent.open.create_mode = 0;
1164 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1165 if (IS_ERR(nd->intent.open.file)) {
1167 err = PTR_ERR(nd->intent.open.file);
1168 path_put(&nd->path);
1170 } else if (err != 0)
1171 release_open_intent(nd);
1175 static struct dentry *__lookup_hash(struct qstr *name,
1176 struct dentry *base, struct nameidata *nd)
1178 struct dentry *dentry;
1179 struct inode *inode;
1182 inode = base->d_inode;
1185 * See if the low-level filesystem might want
1186 * to use its own hash..
1188 if (base->d_op && base->d_op->d_hash) {
1189 err = base->d_op->d_hash(base, name);
1190 dentry = ERR_PTR(err);
1195 dentry = cached_lookup(base, name, nd);
1199 /* Don't create child dentry for a dead directory. */
1200 dentry = ERR_PTR(-ENOENT);
1201 if (IS_DEADDIR(inode))
1204 new = d_alloc(base, name);
1205 dentry = ERR_PTR(-ENOMEM);
1208 dentry = inode->i_op->lookup(inode, new, nd);
1219 * Restricted form of lookup. Doesn't follow links, single-component only,
1220 * needs parent already locked. Doesn't follow mounts.
1223 static struct dentry *lookup_hash(struct nameidata *nd)
1227 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1229 return ERR_PTR(err);
1230 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1233 static int __lookup_one_len(const char *name, struct qstr *this,
1234 struct dentry *base, int len)
1244 hash = init_name_hash();
1246 c = *(const unsigned char *)name++;
1247 if (c == '/' || c == '\0')
1249 hash = partial_name_hash(c, hash);
1251 this->hash = end_name_hash(hash);
1256 * lookup_one_len - filesystem helper to lookup single pathname component
1257 * @name: pathname component to lookup
1258 * @base: base directory to lookup from
1259 * @len: maximum length @len should be interpreted to
1261 * Note that this routine is purely a helper for filesystem usage and should
1262 * not be called by generic code. Also note that by using this function the
1263 * nameidata argument is passed to the filesystem methods and a filesystem
1264 * using this helper needs to be prepared for that.
1266 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1271 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1273 err = __lookup_one_len(name, &this, base, len);
1275 return ERR_PTR(err);
1277 err = inode_permission(base->d_inode, MAY_EXEC);
1279 return ERR_PTR(err);
1280 return __lookup_hash(&this, base, NULL);
1284 * lookup_one_noperm - bad hack for sysfs
1285 * @name: pathname component to lookup
1286 * @base: base directory to lookup from
1288 * This is a variant of lookup_one_len that doesn't perform any permission
1289 * checks. It's a horrible hack to work around the braindead sysfs
1290 * architecture and should not be used anywhere else.
1292 * DON'T USE THIS FUNCTION EVER, thanks.
1294 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1299 err = __lookup_one_len(name, &this, base, strlen(name));
1301 return ERR_PTR(err);
1302 return __lookup_hash(&this, base, NULL);
1305 int user_path_at(int dfd, const char __user *name, unsigned flags,
1308 struct nameidata nd;
1309 char *tmp = getname(name);
1310 int err = PTR_ERR(tmp);
1313 BUG_ON(flags & LOOKUP_PARENT);
1315 err = do_path_lookup(dfd, tmp, flags, &nd);
1323 static int user_path_parent(int dfd, const char __user *path,
1324 struct nameidata *nd, char **name)
1326 char *s = getname(path);
1332 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1342 * It's inline, so penalty for filesystems that don't use sticky bit is
1345 static inline int check_sticky(struct inode *dir, struct inode *inode)
1347 uid_t fsuid = current_fsuid();
1349 if (!(dir->i_mode & S_ISVTX))
1351 if (inode->i_uid == fsuid)
1353 if (dir->i_uid == fsuid)
1355 return !capable(CAP_FOWNER);
1359 * Check whether we can remove a link victim from directory dir, check
1360 * whether the type of victim is right.
1361 * 1. We can't do it if dir is read-only (done in permission())
1362 * 2. We should have write and exec permissions on dir
1363 * 3. We can't remove anything from append-only dir
1364 * 4. We can't do anything with immutable dir (done in permission())
1365 * 5. If the sticky bit on dir is set we should either
1366 * a. be owner of dir, or
1367 * b. be owner of victim, or
1368 * c. have CAP_FOWNER capability
1369 * 6. If the victim is append-only or immutable we can't do antyhing with
1370 * links pointing to it.
1371 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1372 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1373 * 9. We can't remove a root or mountpoint.
1374 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1375 * nfs_async_unlink().
1377 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1381 if (!victim->d_inode)
1384 BUG_ON(victim->d_parent->d_inode != dir);
1385 audit_inode_child(victim->d_name.name, victim, dir);
1387 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1392 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1393 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1396 if (!S_ISDIR(victim->d_inode->i_mode))
1398 if (IS_ROOT(victim))
1400 } else if (S_ISDIR(victim->d_inode->i_mode))
1402 if (IS_DEADDIR(dir))
1404 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1409 /* Check whether we can create an object with dentry child in directory
1411 * 1. We can't do it if child already exists (open has special treatment for
1412 * this case, but since we are inlined it's OK)
1413 * 2. We can't do it if dir is read-only (done in permission())
1414 * 3. We should have write and exec permissions on dir
1415 * 4. We can't do it if dir is immutable (done in permission())
1417 static inline int may_create(struct inode *dir, struct dentry *child)
1421 if (IS_DEADDIR(dir))
1423 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1427 * O_DIRECTORY translates into forcing a directory lookup.
1429 static inline int lookup_flags(unsigned int f)
1431 unsigned long retval = LOOKUP_FOLLOW;
1434 retval &= ~LOOKUP_FOLLOW;
1436 if (f & O_DIRECTORY)
1437 retval |= LOOKUP_DIRECTORY;
1443 * p1 and p2 should be directories on the same fs.
1445 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1450 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1454 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1456 p = d_ancestor(p2, p1);
1458 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1459 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1463 p = d_ancestor(p1, p2);
1465 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1466 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1470 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1471 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1475 void unlock_rename(struct dentry *p1, struct dentry *p2)
1477 mutex_unlock(&p1->d_inode->i_mutex);
1479 mutex_unlock(&p2->d_inode->i_mutex);
1480 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1484 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1485 struct nameidata *nd)
1487 int error = may_create(dir, dentry);
1492 if (!dir->i_op->create)
1493 return -EACCES; /* shouldn't it be ENOSYS? */
1496 error = security_inode_create(dir, dentry, mode);
1500 error = dir->i_op->create(dir, dentry, mode, nd);
1502 fsnotify_create(dir, dentry);
1506 int may_open(struct path *path, int acc_mode, int flag)
1508 struct dentry *dentry = path->dentry;
1509 struct inode *inode = dentry->d_inode;
1515 switch (inode->i_mode & S_IFMT) {
1519 if (acc_mode & MAY_WRITE)
1524 if (path->mnt->mnt_flags & MNT_NODEV)
1533 error = inode_permission(inode, acc_mode);
1537 error = ima_path_check(path, acc_mode ?
1538 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1539 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1545 * An append-only file must be opened in append mode for writing.
1547 if (IS_APPEND(inode)) {
1549 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1555 /* O_NOATIME can only be set by the owner or superuser */
1556 if (flag & O_NOATIME)
1557 if (!is_owner_or_cap(inode)) {
1563 * Ensure there are no outstanding leases on the file.
1565 error = break_lease(inode, flag);
1569 if (flag & O_TRUNC) {
1570 error = get_write_access(inode);
1575 * Refuse to truncate files with mandatory locks held on them.
1577 error = locks_verify_locked(inode);
1579 error = security_path_truncate(path, 0,
1580 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1584 error = do_truncate(dentry, 0,
1585 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1588 put_write_access(inode);
1592 if (flag & FMODE_WRITE)
1597 ima_counts_put(path, acc_mode ?
1598 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1599 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1604 * Be careful about ever adding any more callers of this
1605 * function. Its flags must be in the namei format, not
1606 * what get passed to sys_open().
1608 static int __open_namei_create(struct nameidata *nd, struct path *path,
1612 struct dentry *dir = nd->path.dentry;
1614 if (!IS_POSIXACL(dir->d_inode))
1615 mode &= ~current_umask();
1616 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1619 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1621 mutex_unlock(&dir->d_inode->i_mutex);
1622 dput(nd->path.dentry);
1623 nd->path.dentry = path->dentry;
1626 /* Don't check for write permission, don't truncate */
1627 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1631 * Note that while the flag value (low two bits) for sys_open means:
1636 * it is changed into
1637 * 00 - no permissions needed
1638 * 01 - read-permission
1639 * 10 - write-permission
1641 * for the internal routines (ie open_namei()/follow_link() etc)
1642 * This is more logical, and also allows the 00 "no perm needed"
1643 * to be used for symlinks (where the permissions are checked
1647 static inline int open_to_namei_flags(int flag)
1649 if ((flag+1) & O_ACCMODE)
1654 static int open_will_write_to_fs(int flag, struct inode *inode)
1657 * We'll never write to the fs underlying
1660 if (special_file(inode->i_mode))
1662 return (flag & O_TRUNC);
1666 * Note that the low bits of the passed in "open_flag"
1667 * are not the same as in the local variable "flag". See
1668 * open_to_namei_flags() for more details.
1670 struct file *do_filp_open(int dfd, const char *pathname,
1671 int open_flag, int mode, int acc_mode)
1674 struct nameidata nd;
1680 int flag = open_to_namei_flags(open_flag);
1683 acc_mode = MAY_OPEN | ACC_MODE(flag);
1685 /* O_TRUNC implies we need access checks for write permissions */
1687 acc_mode |= MAY_WRITE;
1689 /* Allow the LSM permission hook to distinguish append
1690 access from general write access. */
1691 if (flag & O_APPEND)
1692 acc_mode |= MAY_APPEND;
1695 * The simplest case - just a plain lookup.
1697 if (!(flag & O_CREAT)) {
1698 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1701 return ERR_PTR(error);
1706 * Create - we need to know the parent.
1708 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1710 return ERR_PTR(error);
1711 error = path_walk(pathname, &nd);
1715 return ERR_PTR(error);
1717 if (unlikely(!audit_dummy_context()))
1718 audit_inode(pathname, nd.path.dentry);
1721 * We have the parent and last component. First of all, check
1722 * that we are not asked to creat(2) an obvious directory - that
1726 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1730 filp = get_empty_filp();
1733 nd.intent.open.file = filp;
1734 nd.intent.open.flags = flag;
1735 nd.intent.open.create_mode = mode;
1736 dir = nd.path.dentry;
1737 nd.flags &= ~LOOKUP_PARENT;
1738 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1740 nd.flags |= LOOKUP_EXCL;
1741 mutex_lock(&dir->d_inode->i_mutex);
1742 path.dentry = lookup_hash(&nd);
1743 path.mnt = nd.path.mnt;
1746 error = PTR_ERR(path.dentry);
1747 if (IS_ERR(path.dentry)) {
1748 mutex_unlock(&dir->d_inode->i_mutex);
1752 if (IS_ERR(nd.intent.open.file)) {
1753 error = PTR_ERR(nd.intent.open.file);
1754 goto exit_mutex_unlock;
1757 /* Negative dentry, just create the file */
1758 if (!path.dentry->d_inode) {
1760 * This write is needed to ensure that a
1761 * ro->rw transition does not occur between
1762 * the time when the file is created and when
1763 * a permanent write count is taken through
1764 * the 'struct file' in nameidata_to_filp().
1766 error = mnt_want_write(nd.path.mnt);
1768 goto exit_mutex_unlock;
1769 error = __open_namei_create(&nd, &path, flag, mode);
1771 mnt_drop_write(nd.path.mnt);
1774 filp = nameidata_to_filp(&nd, open_flag);
1776 ima_counts_put(&nd.path,
1777 acc_mode & (MAY_READ | MAY_WRITE |
1779 mnt_drop_write(nd.path.mnt);
1786 * It already exists.
1788 mutex_unlock(&dir->d_inode->i_mutex);
1789 audit_inode(pathname, path.dentry);
1795 if (__follow_mount(&path)) {
1797 if (flag & O_NOFOLLOW)
1802 if (!path.dentry->d_inode)
1804 if (path.dentry->d_inode->i_op->follow_link)
1807 path_to_nameidata(&path, &nd);
1809 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1814 * 1. may_open() truncates a file
1815 * 2. a rw->ro mount transition occurs
1816 * 3. nameidata_to_filp() fails due to
1818 * That would be inconsistent, and should
1819 * be avoided. Taking this mnt write here
1820 * ensures that (2) can not occur.
1822 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1824 error = mnt_want_write(nd.path.mnt);
1828 error = may_open(&nd.path, acc_mode, flag);
1831 mnt_drop_write(nd.path.mnt);
1834 filp = nameidata_to_filp(&nd, open_flag);
1836 ima_counts_put(&nd.path,
1837 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1839 * It is now safe to drop the mnt write
1840 * because the filp has had a write taken
1844 mnt_drop_write(nd.path.mnt);
1850 mutex_unlock(&dir->d_inode->i_mutex);
1852 path_put_conditional(&path, &nd);
1854 if (!IS_ERR(nd.intent.open.file))
1855 release_open_intent(&nd);
1860 return ERR_PTR(error);
1864 if (flag & O_NOFOLLOW)
1867 * This is subtle. Instead of calling do_follow_link() we do the
1868 * thing by hands. The reason is that this way we have zero link_count
1869 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1870 * After that we have the parent and last component, i.e.
1871 * we are in the same situation as after the first path_walk().
1872 * Well, almost - if the last component is normal we get its copy
1873 * stored in nd->last.name and we will have to putname() it when we
1874 * are done. Procfs-like symlinks just set LAST_BIND.
1876 nd.flags |= LOOKUP_PARENT;
1877 error = security_inode_follow_link(path.dentry, &nd);
1880 error = __do_follow_link(&path, &nd);
1882 /* Does someone understand code flow here? Or it is only
1883 * me so stupid? Anathema to whoever designed this non-sense
1884 * with "intent.open".
1886 release_open_intent(&nd);
1889 return ERR_PTR(error);
1891 nd.flags &= ~LOOKUP_PARENT;
1892 if (nd.last_type == LAST_BIND)
1895 if (nd.last_type != LAST_NORM)
1897 if (nd.last.name[nd.last.len]) {
1898 __putname(nd.last.name);
1903 __putname(nd.last.name);
1906 dir = nd.path.dentry;
1907 mutex_lock(&dir->d_inode->i_mutex);
1908 path.dentry = lookup_hash(&nd);
1909 path.mnt = nd.path.mnt;
1910 __putname(nd.last.name);
1915 * filp_open - open file and return file pointer
1917 * @filename: path to open
1918 * @flags: open flags as per the open(2) second argument
1919 * @mode: mode for the new file if O_CREAT is set, else ignored
1921 * This is the helper to open a file from kernelspace if you really
1922 * have to. But in generally you should not do this, so please move
1923 * along, nothing to see here..
1925 struct file *filp_open(const char *filename, int flags, int mode)
1927 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1929 EXPORT_SYMBOL(filp_open);
1932 * lookup_create - lookup a dentry, creating it if it doesn't exist
1933 * @nd: nameidata info
1934 * @is_dir: directory flag
1936 * Simple function to lookup and return a dentry and create it
1937 * if it doesn't exist. Is SMP-safe.
1939 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1941 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1943 struct dentry *dentry = ERR_PTR(-EEXIST);
1945 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1947 * Yucky last component or no last component at all?
1948 * (foo/., foo/.., /////)
1950 if (nd->last_type != LAST_NORM)
1952 nd->flags &= ~LOOKUP_PARENT;
1953 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1954 nd->intent.open.flags = O_EXCL;
1957 * Do the final lookup.
1959 dentry = lookup_hash(nd);
1963 if (dentry->d_inode)
1966 * Special case - lookup gave negative, but... we had foo/bar/
1967 * From the vfs_mknod() POV we just have a negative dentry -
1968 * all is fine. Let's be bastards - you had / on the end, you've
1969 * been asking for (non-existent) directory. -ENOENT for you.
1971 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1973 dentry = ERR_PTR(-ENOENT);
1978 dentry = ERR_PTR(-EEXIST);
1982 EXPORT_SYMBOL_GPL(lookup_create);
1984 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1986 int error = may_create(dir, dentry);
1991 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1994 if (!dir->i_op->mknod)
1997 error = devcgroup_inode_mknod(mode, dev);
2001 error = security_inode_mknod(dir, dentry, mode, dev);
2006 error = dir->i_op->mknod(dir, dentry, mode, dev);
2008 fsnotify_create(dir, dentry);
2012 static int may_mknod(mode_t mode)
2014 switch (mode & S_IFMT) {
2020 case 0: /* zero mode translates to S_IFREG */
2029 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2034 struct dentry *dentry;
2035 struct nameidata nd;
2040 error = user_path_parent(dfd, filename, &nd, &tmp);
2044 dentry = lookup_create(&nd, 0);
2045 if (IS_ERR(dentry)) {
2046 error = PTR_ERR(dentry);
2049 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2050 mode &= ~current_umask();
2051 error = may_mknod(mode);
2054 error = mnt_want_write(nd.path.mnt);
2057 error = security_path_mknod(&nd.path, dentry, mode, dev);
2059 goto out_drop_write;
2060 switch (mode & S_IFMT) {
2061 case 0: case S_IFREG:
2062 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2064 case S_IFCHR: case S_IFBLK:
2065 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2066 new_decode_dev(dev));
2068 case S_IFIFO: case S_IFSOCK:
2069 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2073 mnt_drop_write(nd.path.mnt);
2077 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2084 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2086 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2089 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2091 int error = may_create(dir, dentry);
2096 if (!dir->i_op->mkdir)
2099 mode &= (S_IRWXUGO|S_ISVTX);
2100 error = security_inode_mkdir(dir, dentry, mode);
2105 error = dir->i_op->mkdir(dir, dentry, mode);
2107 fsnotify_mkdir(dir, dentry);
2111 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2115 struct dentry *dentry;
2116 struct nameidata nd;
2118 error = user_path_parent(dfd, pathname, &nd, &tmp);
2122 dentry = lookup_create(&nd, 1);
2123 error = PTR_ERR(dentry);
2127 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2128 mode &= ~current_umask();
2129 error = mnt_want_write(nd.path.mnt);
2132 error = security_path_mkdir(&nd.path, dentry, mode);
2134 goto out_drop_write;
2135 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2137 mnt_drop_write(nd.path.mnt);
2141 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2148 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2150 return sys_mkdirat(AT_FDCWD, pathname, mode);
2154 * We try to drop the dentry early: we should have
2155 * a usage count of 2 if we're the only user of this
2156 * dentry, and if that is true (possibly after pruning
2157 * the dcache), then we drop the dentry now.
2159 * A low-level filesystem can, if it choses, legally
2162 * if (!d_unhashed(dentry))
2165 * if it cannot handle the case of removing a directory
2166 * that is still in use by something else..
2168 void dentry_unhash(struct dentry *dentry)
2171 shrink_dcache_parent(dentry);
2172 spin_lock(&dcache_lock);
2173 spin_lock(&dentry->d_lock);
2174 if (atomic_read(&dentry->d_count) == 2)
2176 spin_unlock(&dentry->d_lock);
2177 spin_unlock(&dcache_lock);
2180 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2182 int error = may_delete(dir, dentry, 1);
2187 if (!dir->i_op->rmdir)
2192 mutex_lock(&dentry->d_inode->i_mutex);
2193 dentry_unhash(dentry);
2194 if (d_mountpoint(dentry))
2197 error = security_inode_rmdir(dir, dentry);
2199 error = dir->i_op->rmdir(dir, dentry);
2201 dentry->d_inode->i_flags |= S_DEAD;
2204 mutex_unlock(&dentry->d_inode->i_mutex);
2213 static long do_rmdir(int dfd, const char __user *pathname)
2217 struct dentry *dentry;
2218 struct nameidata nd;
2220 error = user_path_parent(dfd, pathname, &nd, &name);
2224 switch(nd.last_type) {
2236 nd.flags &= ~LOOKUP_PARENT;
2238 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2239 dentry = lookup_hash(&nd);
2240 error = PTR_ERR(dentry);
2243 error = mnt_want_write(nd.path.mnt);
2246 error = security_path_rmdir(&nd.path, dentry);
2249 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2251 mnt_drop_write(nd.path.mnt);
2255 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2262 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2264 return do_rmdir(AT_FDCWD, pathname);
2267 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2269 int error = may_delete(dir, dentry, 0);
2274 if (!dir->i_op->unlink)
2279 mutex_lock(&dentry->d_inode->i_mutex);
2280 if (d_mountpoint(dentry))
2283 error = security_inode_unlink(dir, dentry);
2285 error = dir->i_op->unlink(dir, dentry);
2287 mutex_unlock(&dentry->d_inode->i_mutex);
2289 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2290 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2291 fsnotify_link_count(dentry->d_inode);
2299 * Make sure that the actual truncation of the file will occur outside its
2300 * directory's i_mutex. Truncate can take a long time if there is a lot of
2301 * writeout happening, and we don't want to prevent access to the directory
2302 * while waiting on the I/O.
2304 static long do_unlinkat(int dfd, const char __user *pathname)
2308 struct dentry *dentry;
2309 struct nameidata nd;
2310 struct inode *inode = NULL;
2312 error = user_path_parent(dfd, pathname, &nd, &name);
2317 if (nd.last_type != LAST_NORM)
2320 nd.flags &= ~LOOKUP_PARENT;
2322 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2323 dentry = lookup_hash(&nd);
2324 error = PTR_ERR(dentry);
2325 if (!IS_ERR(dentry)) {
2326 /* Why not before? Because we want correct error value */
2327 if (nd.last.name[nd.last.len])
2329 inode = dentry->d_inode;
2331 atomic_inc(&inode->i_count);
2332 error = mnt_want_write(nd.path.mnt);
2335 error = security_path_unlink(&nd.path, dentry);
2338 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2340 mnt_drop_write(nd.path.mnt);
2344 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2346 iput(inode); /* truncate the inode here */
2353 error = !dentry->d_inode ? -ENOENT :
2354 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2358 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2360 if ((flag & ~AT_REMOVEDIR) != 0)
2363 if (flag & AT_REMOVEDIR)
2364 return do_rmdir(dfd, pathname);
2366 return do_unlinkat(dfd, pathname);
2369 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2371 return do_unlinkat(AT_FDCWD, pathname);
2374 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2376 int error = may_create(dir, dentry);
2381 if (!dir->i_op->symlink)
2384 error = security_inode_symlink(dir, dentry, oldname);
2389 error = dir->i_op->symlink(dir, dentry, oldname);
2391 fsnotify_create(dir, dentry);
2395 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2396 int, newdfd, const char __user *, newname)
2401 struct dentry *dentry;
2402 struct nameidata nd;
2404 from = getname(oldname);
2406 return PTR_ERR(from);
2408 error = user_path_parent(newdfd, newname, &nd, &to);
2412 dentry = lookup_create(&nd, 0);
2413 error = PTR_ERR(dentry);
2417 error = mnt_want_write(nd.path.mnt);
2420 error = security_path_symlink(&nd.path, dentry, from);
2422 goto out_drop_write;
2423 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2425 mnt_drop_write(nd.path.mnt);
2429 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2437 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2439 return sys_symlinkat(oldname, AT_FDCWD, newname);
2442 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2444 struct inode *inode = old_dentry->d_inode;
2450 error = may_create(dir, new_dentry);
2454 if (dir->i_sb != inode->i_sb)
2458 * A link to an append-only or immutable file cannot be created.
2460 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2462 if (!dir->i_op->link)
2464 if (S_ISDIR(inode->i_mode))
2467 error = security_inode_link(old_dentry, dir, new_dentry);
2471 mutex_lock(&inode->i_mutex);
2473 error = dir->i_op->link(old_dentry, dir, new_dentry);
2474 mutex_unlock(&inode->i_mutex);
2476 fsnotify_link(dir, inode, new_dentry);
2481 * Hardlinks are often used in delicate situations. We avoid
2482 * security-related surprises by not following symlinks on the
2485 * We don't follow them on the oldname either to be compatible
2486 * with linux 2.0, and to avoid hard-linking to directories
2487 * and other special files. --ADM
2489 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2490 int, newdfd, const char __user *, newname, int, flags)
2492 struct dentry *new_dentry;
2493 struct nameidata nd;
2494 struct path old_path;
2498 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2501 error = user_path_at(olddfd, oldname,
2502 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2507 error = user_path_parent(newdfd, newname, &nd, &to);
2511 if (old_path.mnt != nd.path.mnt)
2513 new_dentry = lookup_create(&nd, 0);
2514 error = PTR_ERR(new_dentry);
2515 if (IS_ERR(new_dentry))
2517 error = mnt_want_write(nd.path.mnt);
2520 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2522 goto out_drop_write;
2523 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2525 mnt_drop_write(nd.path.mnt);
2529 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2534 path_put(&old_path);
2539 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2541 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2545 * The worst of all namespace operations - renaming directory. "Perverted"
2546 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2548 * a) we can get into loop creation. Check is done in is_subdir().
2549 * b) race potential - two innocent renames can create a loop together.
2550 * That's where 4.4 screws up. Current fix: serialization on
2551 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2553 * c) we have to lock _three_ objects - parents and victim (if it exists).
2554 * And that - after we got ->i_mutex on parents (until then we don't know
2555 * whether the target exists). Solution: try to be smart with locking
2556 * order for inodes. We rely on the fact that tree topology may change
2557 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2558 * move will be locked. Thus we can rank directories by the tree
2559 * (ancestors first) and rank all non-directories after them.
2560 * That works since everybody except rename does "lock parent, lookup,
2561 * lock child" and rename is under ->s_vfs_rename_mutex.
2562 * HOWEVER, it relies on the assumption that any object with ->lookup()
2563 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2564 * we'd better make sure that there's no link(2) for them.
2565 * d) some filesystems don't support opened-but-unlinked directories,
2566 * either because of layout or because they are not ready to deal with
2567 * all cases correctly. The latter will be fixed (taking this sort of
2568 * stuff into VFS), but the former is not going away. Solution: the same
2569 * trick as in rmdir().
2570 * e) conversion from fhandle to dentry may come in the wrong moment - when
2571 * we are removing the target. Solution: we will have to grab ->i_mutex
2572 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2573 * ->i_mutex on parents, which works but leads to some truely excessive
2576 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2577 struct inode *new_dir, struct dentry *new_dentry)
2580 struct inode *target;
2583 * If we are going to change the parent - check write permissions,
2584 * we'll need to flip '..'.
2586 if (new_dir != old_dir) {
2587 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2592 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2596 target = new_dentry->d_inode;
2598 mutex_lock(&target->i_mutex);
2599 dentry_unhash(new_dentry);
2601 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2604 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2607 target->i_flags |= S_DEAD;
2608 mutex_unlock(&target->i_mutex);
2609 if (d_unhashed(new_dentry))
2610 d_rehash(new_dentry);
2614 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2615 d_move(old_dentry,new_dentry);
2619 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2620 struct inode *new_dir, struct dentry *new_dentry)
2622 struct inode *target;
2625 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2630 target = new_dentry->d_inode;
2632 mutex_lock(&target->i_mutex);
2633 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2636 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2638 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2639 d_move(old_dentry, new_dentry);
2642 mutex_unlock(&target->i_mutex);
2647 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2648 struct inode *new_dir, struct dentry *new_dentry)
2651 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2652 const char *old_name;
2654 if (old_dentry->d_inode == new_dentry->d_inode)
2657 error = may_delete(old_dir, old_dentry, is_dir);
2661 if (!new_dentry->d_inode)
2662 error = may_create(new_dir, new_dentry);
2664 error = may_delete(new_dir, new_dentry, is_dir);
2668 if (!old_dir->i_op->rename)
2671 vfs_dq_init(old_dir);
2672 vfs_dq_init(new_dir);
2674 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2677 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2679 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2681 const char *new_name = old_dentry->d_name.name;
2682 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2683 new_dentry->d_inode, old_dentry);
2685 fsnotify_oldname_free(old_name);
2690 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2691 int, newdfd, const char __user *, newname)
2693 struct dentry *old_dir, *new_dir;
2694 struct dentry *old_dentry, *new_dentry;
2695 struct dentry *trap;
2696 struct nameidata oldnd, newnd;
2701 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2705 error = user_path_parent(newdfd, newname, &newnd, &to);
2710 if (oldnd.path.mnt != newnd.path.mnt)
2713 old_dir = oldnd.path.dentry;
2715 if (oldnd.last_type != LAST_NORM)
2718 new_dir = newnd.path.dentry;
2719 if (newnd.last_type != LAST_NORM)
2722 oldnd.flags &= ~LOOKUP_PARENT;
2723 newnd.flags &= ~LOOKUP_PARENT;
2724 newnd.flags |= LOOKUP_RENAME_TARGET;
2726 trap = lock_rename(new_dir, old_dir);
2728 old_dentry = lookup_hash(&oldnd);
2729 error = PTR_ERR(old_dentry);
2730 if (IS_ERR(old_dentry))
2732 /* source must exist */
2734 if (!old_dentry->d_inode)
2736 /* unless the source is a directory trailing slashes give -ENOTDIR */
2737 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2739 if (oldnd.last.name[oldnd.last.len])
2741 if (newnd.last.name[newnd.last.len])
2744 /* source should not be ancestor of target */
2746 if (old_dentry == trap)
2748 new_dentry = lookup_hash(&newnd);
2749 error = PTR_ERR(new_dentry);
2750 if (IS_ERR(new_dentry))
2752 /* target should not be an ancestor of source */
2754 if (new_dentry == trap)
2757 error = mnt_want_write(oldnd.path.mnt);
2760 error = security_path_rename(&oldnd.path, old_dentry,
2761 &newnd.path, new_dentry);
2764 error = vfs_rename(old_dir->d_inode, old_dentry,
2765 new_dir->d_inode, new_dentry);
2767 mnt_drop_write(oldnd.path.mnt);
2773 unlock_rename(new_dir, old_dir);
2775 path_put(&newnd.path);
2778 path_put(&oldnd.path);
2784 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2786 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2789 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2793 len = PTR_ERR(link);
2798 if (len > (unsigned) buflen)
2800 if (copy_to_user(buffer, link, len))
2807 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2808 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2809 * using) it for any given inode is up to filesystem.
2811 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2813 struct nameidata nd;
2818 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2820 return PTR_ERR(cookie);
2822 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2823 if (dentry->d_inode->i_op->put_link)
2824 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2828 int vfs_follow_link(struct nameidata *nd, const char *link)
2830 return __vfs_follow_link(nd, link);
2833 /* get the link contents into pagecache */
2834 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2838 struct address_space *mapping = dentry->d_inode->i_mapping;
2839 page = read_mapping_page(mapping, 0, NULL);
2844 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2848 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2850 struct page *page = NULL;
2851 char *s = page_getlink(dentry, &page);
2852 int res = vfs_readlink(dentry,buffer,buflen,s);
2855 page_cache_release(page);
2860 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2862 struct page *page = NULL;
2863 nd_set_link(nd, page_getlink(dentry, &page));
2867 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2869 struct page *page = cookie;
2873 page_cache_release(page);
2878 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2880 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2882 struct address_space *mapping = inode->i_mapping;
2887 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2889 flags |= AOP_FLAG_NOFS;
2892 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2893 flags, &page, &fsdata);
2897 kaddr = kmap_atomic(page, KM_USER0);
2898 memcpy(kaddr, symname, len-1);
2899 kunmap_atomic(kaddr, KM_USER0);
2901 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2908 mark_inode_dirty(inode);
2914 int page_symlink(struct inode *inode, const char *symname, int len)
2916 return __page_symlink(inode, symname, len,
2917 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2920 const struct inode_operations page_symlink_inode_operations = {
2921 .readlink = generic_readlink,
2922 .follow_link = page_follow_link_light,
2923 .put_link = page_put_link,
2926 EXPORT_SYMBOL(user_path_at);
2927 EXPORT_SYMBOL(follow_down);
2928 EXPORT_SYMBOL(follow_up);
2929 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2930 EXPORT_SYMBOL(getname);
2931 EXPORT_SYMBOL(lock_rename);
2932 EXPORT_SYMBOL(lookup_one_len);
2933 EXPORT_SYMBOL(page_follow_link_light);
2934 EXPORT_SYMBOL(page_put_link);
2935 EXPORT_SYMBOL(page_readlink);
2936 EXPORT_SYMBOL(__page_symlink);
2937 EXPORT_SYMBOL(page_symlink);
2938 EXPORT_SYMBOL(page_symlink_inode_operations);
2939 EXPORT_SYMBOL(path_lookup);
2940 EXPORT_SYMBOL(kern_path);
2941 EXPORT_SYMBOL(vfs_path_lookup);
2942 EXPORT_SYMBOL(inode_permission);
2943 EXPORT_SYMBOL(file_permission);
2944 EXPORT_SYMBOL(unlock_rename);
2945 EXPORT_SYMBOL(vfs_create);
2946 EXPORT_SYMBOL(vfs_follow_link);
2947 EXPORT_SYMBOL(vfs_link);
2948 EXPORT_SYMBOL(vfs_mkdir);
2949 EXPORT_SYMBOL(vfs_mknod);
2950 EXPORT_SYMBOL(generic_permission);
2951 EXPORT_SYMBOL(vfs_readlink);
2952 EXPORT_SYMBOL(vfs_rename);
2953 EXPORT_SYMBOL(vfs_rmdir);
2954 EXPORT_SYMBOL(vfs_symlink);
2955 EXPORT_SYMBOL(vfs_unlink);
2956 EXPORT_SYMBOL(dentry_unhash);
2957 EXPORT_SYMBOL(generic_readlink);