Merge tag 'for-linus-20130331' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / fs / namei.c
blob57ae9c8c66bfc6d98ae96643874710cdc3de321a
1 /*
2 * linux/fs/namei.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
8 * Some corrections by tytso.
9 */
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
21 #include <linux/fs.h>
22 #include <linux/namei.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 <linux/posix_acl.h>
37 #include <asm/uaccess.h>
39 #include "internal.h"
40 #include "mount.h"
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
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 void final_putname(struct filename *name)
122 if (name->separate) {
123 __putname(name->name);
124 kfree(name);
125 } else {
126 __putname(name);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename *
133 getname_flags(const char __user *filename, int flags, int *empty)
135 struct filename *result, *err;
136 int len;
137 long max;
138 char *kname;
140 result = audit_reusename(filename);
141 if (result)
142 return result;
144 result = __getname();
145 if (unlikely(!result))
146 return ERR_PTR(-ENOMEM);
149 * First, try to embed the struct filename inside the names_cache
150 * allocation
152 kname = (char *)result + sizeof(*result);
153 result->name = kname;
154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
157 recopy:
158 len = strncpy_from_user(kname, filename, max);
159 if (unlikely(len < 0)) {
160 err = ERR_PTR(len);
161 goto error;
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
168 * userland.
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
174 if (!result) {
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
177 goto error;
179 result->name = kname;
180 result->separate = true;
181 max = PATH_MAX;
182 goto recopy;
185 /* The empty path is special. */
186 if (unlikely(!len)) {
187 if (empty)
188 *empty = 1;
189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
191 goto error;
194 err = ERR_PTR(-ENAMETOOLONG);
195 if (unlikely(len >= PATH_MAX))
196 goto error;
198 result->uptr = filename;
199 audit_getname(result);
200 return result;
202 error:
203 final_putname(result);
204 return err;
207 struct filename *
208 getname(const char __user * filename)
210 return getname_flags(filename, 0, NULL);
212 EXPORT_SYMBOL(getname);
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename *name)
217 if (unlikely(!audit_dummy_context()))
218 return audit_putname(name);
219 final_putname(name);
221 #endif
223 static int check_acl(struct inode *inode, int mask)
225 #ifdef CONFIG_FS_POSIX_ACL
226 struct posix_acl *acl;
228 if (mask & MAY_NOT_BLOCK) {
229 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
230 if (!acl)
231 return -EAGAIN;
232 /* no ->get_acl() calls in RCU mode... */
233 if (acl == ACL_NOT_CACHED)
234 return -ECHILD;
235 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
238 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
241 * A filesystem can force a ACL callback by just never filling the
242 * ACL cache. But normally you'd fill the cache either at inode
243 * instantiation time, or on the first ->get_acl call.
245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
248 if (acl == ACL_NOT_CACHED) {
249 if (inode->i_op->get_acl) {
250 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
251 if (IS_ERR(acl))
252 return PTR_ERR(acl);
253 } else {
254 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
255 return -EAGAIN;
259 if (acl) {
260 int error = posix_acl_permission(inode, acl, mask);
261 posix_acl_release(acl);
262 return error;
264 #endif
266 return -EAGAIN;
270 * This does the basic permission checking
272 static int acl_permission_check(struct inode *inode, int mask)
274 unsigned int mode = inode->i_mode;
276 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
277 mode >>= 6;
278 else {
279 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
280 int error = check_acl(inode, mask);
281 if (error != -EAGAIN)
282 return error;
285 if (in_group_p(inode->i_gid))
286 mode >>= 3;
290 * If the DACs are ok we don't need any capability check.
292 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
293 return 0;
294 return -EACCES;
298 * generic_permission - check for access rights on a Posix-like filesystem
299 * @inode: inode to check access rights for
300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
302 * Used to check for read/write/execute permissions on a file.
303 * We use "fsuid" for this, letting us set arbitrary permissions
304 * for filesystem access without changing the "normal" uids which
305 * are used for other things.
307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308 * request cannot be satisfied (eg. requires blocking or too much complexity).
309 * It would then be called again in ref-walk mode.
311 int generic_permission(struct inode *inode, int mask)
313 int ret;
316 * Do the basic permission checks.
318 ret = acl_permission_check(inode, mask);
319 if (ret != -EACCES)
320 return ret;
322 if (S_ISDIR(inode->i_mode)) {
323 /* DACs are overridable for directories */
324 if (inode_capable(inode, CAP_DAC_OVERRIDE))
325 return 0;
326 if (!(mask & MAY_WRITE))
327 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
328 return 0;
329 return -EACCES;
332 * Read/write DACs are always overridable.
333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
336 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
337 if (inode_capable(inode, CAP_DAC_OVERRIDE))
338 return 0;
341 * Searching includes executable on directories, else just read.
343 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
344 if (mask == MAY_READ)
345 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
346 return 0;
348 return -EACCES;
352 * We _really_ want to just do "generic_permission()" without
353 * even looking at the inode->i_op values. So we keep a cache
354 * flag in inode->i_opflags, that says "this has not special
355 * permission function, use the fast case".
357 static inline int do_inode_permission(struct inode *inode, int mask)
359 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
360 if (likely(inode->i_op->permission))
361 return inode->i_op->permission(inode, mask);
363 /* This gets set once for the inode lifetime */
364 spin_lock(&inode->i_lock);
365 inode->i_opflags |= IOP_FASTPERM;
366 spin_unlock(&inode->i_lock);
368 return generic_permission(inode, mask);
372 * __inode_permission - Check for access rights to a given inode
373 * @inode: Inode to check permission on
374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 * Check for read/write/execute permissions on an inode.
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
383 int __inode_permission(struct inode *inode, int mask)
385 int retval;
387 if (unlikely(mask & MAY_WRITE)) {
389 * Nobody gets write access to an immutable file.
391 if (IS_IMMUTABLE(inode))
392 return -EACCES;
395 retval = do_inode_permission(inode, mask);
396 if (retval)
397 return retval;
399 retval = devcgroup_inode_permission(inode, mask);
400 if (retval)
401 return retval;
403 return security_inode_permission(inode, mask);
407 * sb_permission - Check superblock-level permissions
408 * @sb: Superblock of inode to check permission on
409 * @inode: Inode to check permission on
410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
412 * Separate out file-system wide checks from inode-specific permission checks.
414 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
416 if (unlikely(mask & MAY_WRITE)) {
417 umode_t mode = inode->i_mode;
419 /* Nobody gets write access to a read-only fs. */
420 if ((sb->s_flags & MS_RDONLY) &&
421 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
422 return -EROFS;
424 return 0;
428 * inode_permission - Check for access rights to a given inode
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
433 * this, letting us set arbitrary permissions for filesystem access without
434 * changing the "normal" UIDs which are used for other things.
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
438 int inode_permission(struct inode *inode, int mask)
440 int retval;
442 retval = sb_permission(inode->i_sb, inode, mask);
443 if (retval)
444 return retval;
445 return __inode_permission(inode, mask);
449 * path_get - get a reference to a path
450 * @path: path to get the reference to
452 * Given a path increment the reference count to the dentry and the vfsmount.
454 void path_get(const struct path *path)
456 mntget(path->mnt);
457 dget(path->dentry);
459 EXPORT_SYMBOL(path_get);
462 * path_put - put a reference to a path
463 * @path: path to put the reference to
465 * Given a path decrement the reference count to the dentry and the vfsmount.
467 void path_put(const struct path *path)
469 dput(path->dentry);
470 mntput(path->mnt);
472 EXPORT_SYMBOL(path_put);
475 * Path walking has 2 modes, rcu-walk and ref-walk (see
476 * Documentation/filesystems/path-lookup.txt). In situations when we can't
477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479 * mode. Refcounts are grabbed at the last known good point before rcu-walk
480 * got stuck, so ref-walk may continue from there. If this is not successful
481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
482 * to restart the path walk from the beginning in ref-walk mode.
485 static inline void lock_rcu_walk(void)
487 br_read_lock(&vfsmount_lock);
488 rcu_read_lock();
491 static inline void unlock_rcu_walk(void)
493 rcu_read_unlock();
494 br_read_unlock(&vfsmount_lock);
498 * unlazy_walk - try to switch to ref-walk mode.
499 * @nd: nameidata pathwalk data
500 * @dentry: child of nd->path.dentry or NULL
501 * Returns: 0 on success, -ECHILD on failure
503 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
504 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
505 * @nd or NULL. Must be called from rcu-walk context.
507 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
509 struct fs_struct *fs = current->fs;
510 struct dentry *parent = nd->path.dentry;
511 int want_root = 0;
513 BUG_ON(!(nd->flags & LOOKUP_RCU));
514 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
515 want_root = 1;
516 spin_lock(&fs->lock);
517 if (nd->root.mnt != fs->root.mnt ||
518 nd->root.dentry != fs->root.dentry)
519 goto err_root;
521 spin_lock(&parent->d_lock);
522 if (!dentry) {
523 if (!__d_rcu_to_refcount(parent, nd->seq))
524 goto err_parent;
525 BUG_ON(nd->inode != parent->d_inode);
526 } else {
527 if (dentry->d_parent != parent)
528 goto err_parent;
529 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
530 if (!__d_rcu_to_refcount(dentry, nd->seq))
531 goto err_child;
533 * If the sequence check on the child dentry passed, then
534 * the child has not been removed from its parent. This
535 * means the parent dentry must be valid and able to take
536 * a reference at this point.
538 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
539 BUG_ON(!parent->d_count);
540 parent->d_count++;
541 spin_unlock(&dentry->d_lock);
543 spin_unlock(&parent->d_lock);
544 if (want_root) {
545 path_get(&nd->root);
546 spin_unlock(&fs->lock);
548 mntget(nd->path.mnt);
550 unlock_rcu_walk();
551 nd->flags &= ~LOOKUP_RCU;
552 return 0;
554 err_child:
555 spin_unlock(&dentry->d_lock);
556 err_parent:
557 spin_unlock(&parent->d_lock);
558 err_root:
559 if (want_root)
560 spin_unlock(&fs->lock);
561 return -ECHILD;
564 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
566 return dentry->d_op->d_revalidate(dentry, flags);
570 * complete_walk - successful completion of path walk
571 * @nd: pointer nameidata
573 * If we had been in RCU mode, drop out of it and legitimize nd->path.
574 * Revalidate the final result, unless we'd already done that during
575 * the path walk or the filesystem doesn't ask for it. Return 0 on
576 * success, -error on failure. In case of failure caller does not
577 * need to drop nd->path.
579 static int complete_walk(struct nameidata *nd)
581 struct dentry *dentry = nd->path.dentry;
582 int status;
584 if (nd->flags & LOOKUP_RCU) {
585 nd->flags &= ~LOOKUP_RCU;
586 if (!(nd->flags & LOOKUP_ROOT))
587 nd->root.mnt = NULL;
588 spin_lock(&dentry->d_lock);
589 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
590 spin_unlock(&dentry->d_lock);
591 unlock_rcu_walk();
592 return -ECHILD;
594 BUG_ON(nd->inode != dentry->d_inode);
595 spin_unlock(&dentry->d_lock);
596 mntget(nd->path.mnt);
597 unlock_rcu_walk();
600 if (likely(!(nd->flags & LOOKUP_JUMPED)))
601 return 0;
603 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
604 return 0;
606 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
607 if (status > 0)
608 return 0;
610 if (!status)
611 status = -ESTALE;
613 path_put(&nd->path);
614 return status;
617 static __always_inline void set_root(struct nameidata *nd)
619 if (!nd->root.mnt)
620 get_fs_root(current->fs, &nd->root);
623 static int link_path_walk(const char *, struct nameidata *);
625 static __always_inline void set_root_rcu(struct nameidata *nd)
627 if (!nd->root.mnt) {
628 struct fs_struct *fs = current->fs;
629 unsigned seq;
631 do {
632 seq = read_seqcount_begin(&fs->seq);
633 nd->root = fs->root;
634 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
635 } while (read_seqcount_retry(&fs->seq, seq));
639 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
641 int ret;
643 if (IS_ERR(link))
644 goto fail;
646 if (*link == '/') {
647 set_root(nd);
648 path_put(&nd->path);
649 nd->path = nd->root;
650 path_get(&nd->root);
651 nd->flags |= LOOKUP_JUMPED;
653 nd->inode = nd->path.dentry->d_inode;
655 ret = link_path_walk(link, nd);
656 return ret;
657 fail:
658 path_put(&nd->path);
659 return PTR_ERR(link);
662 static void path_put_conditional(struct path *path, struct nameidata *nd)
664 dput(path->dentry);
665 if (path->mnt != nd->path.mnt)
666 mntput(path->mnt);
669 static inline void path_to_nameidata(const struct path *path,
670 struct nameidata *nd)
672 if (!(nd->flags & LOOKUP_RCU)) {
673 dput(nd->path.dentry);
674 if (nd->path.mnt != path->mnt)
675 mntput(nd->path.mnt);
677 nd->path.mnt = path->mnt;
678 nd->path.dentry = path->dentry;
682 * Helper to directly jump to a known parsed path from ->follow_link,
683 * caller must have taken a reference to path beforehand.
685 void nd_jump_link(struct nameidata *nd, struct path *path)
687 path_put(&nd->path);
689 nd->path = *path;
690 nd->inode = nd->path.dentry->d_inode;
691 nd->flags |= LOOKUP_JUMPED;
694 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
696 struct inode *inode = link->dentry->d_inode;
697 if (inode->i_op->put_link)
698 inode->i_op->put_link(link->dentry, nd, cookie);
699 path_put(link);
702 int sysctl_protected_symlinks __read_mostly = 0;
703 int sysctl_protected_hardlinks __read_mostly = 0;
706 * may_follow_link - Check symlink following for unsafe situations
707 * @link: The path of the symlink
708 * @nd: nameidata pathwalk data
710 * In the case of the sysctl_protected_symlinks sysctl being enabled,
711 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
712 * in a sticky world-writable directory. This is to protect privileged
713 * processes from failing races against path names that may change out
714 * from under them by way of other users creating malicious symlinks.
715 * It will permit symlinks to be followed only when outside a sticky
716 * world-writable directory, or when the uid of the symlink and follower
717 * match, or when the directory owner matches the symlink's owner.
719 * Returns 0 if following the symlink is allowed, -ve on error.
721 static inline int may_follow_link(struct path *link, struct nameidata *nd)
723 const struct inode *inode;
724 const struct inode *parent;
726 if (!sysctl_protected_symlinks)
727 return 0;
729 /* Allowed if owner and follower match. */
730 inode = link->dentry->d_inode;
731 if (uid_eq(current_cred()->fsuid, inode->i_uid))
732 return 0;
734 /* Allowed if parent directory not sticky and world-writable. */
735 parent = nd->path.dentry->d_inode;
736 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
737 return 0;
739 /* Allowed if parent directory and link owner match. */
740 if (uid_eq(parent->i_uid, inode->i_uid))
741 return 0;
743 audit_log_link_denied("follow_link", link);
744 path_put_conditional(link, nd);
745 path_put(&nd->path);
746 return -EACCES;
750 * safe_hardlink_source - Check for safe hardlink conditions
751 * @inode: the source inode to hardlink from
753 * Return false if at least one of the following conditions:
754 * - inode is not a regular file
755 * - inode is setuid
756 * - inode is setgid and group-exec
757 * - access failure for read and write
759 * Otherwise returns true.
761 static bool safe_hardlink_source(struct inode *inode)
763 umode_t mode = inode->i_mode;
765 /* Special files should not get pinned to the filesystem. */
766 if (!S_ISREG(mode))
767 return false;
769 /* Setuid files should not get pinned to the filesystem. */
770 if (mode & S_ISUID)
771 return false;
773 /* Executable setgid files should not get pinned to the filesystem. */
774 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
775 return false;
777 /* Hardlinking to unreadable or unwritable sources is dangerous. */
778 if (inode_permission(inode, MAY_READ | MAY_WRITE))
779 return false;
781 return true;
785 * may_linkat - Check permissions for creating a hardlink
786 * @link: the source to hardlink from
788 * Block hardlink when all of:
789 * - sysctl_protected_hardlinks enabled
790 * - fsuid does not match inode
791 * - hardlink source is unsafe (see safe_hardlink_source() above)
792 * - not CAP_FOWNER
794 * Returns 0 if successful, -ve on error.
796 static int may_linkat(struct path *link)
798 const struct cred *cred;
799 struct inode *inode;
801 if (!sysctl_protected_hardlinks)
802 return 0;
804 cred = current_cred();
805 inode = link->dentry->d_inode;
807 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
808 * otherwise, it must be a safe source.
810 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
811 capable(CAP_FOWNER))
812 return 0;
814 audit_log_link_denied("linkat", link);
815 return -EPERM;
818 static __always_inline int
819 follow_link(struct path *link, struct nameidata *nd, void **p)
821 struct dentry *dentry = link->dentry;
822 int error;
823 char *s;
825 BUG_ON(nd->flags & LOOKUP_RCU);
827 if (link->mnt == nd->path.mnt)
828 mntget(link->mnt);
830 error = -ELOOP;
831 if (unlikely(current->total_link_count >= 40))
832 goto out_put_nd_path;
834 cond_resched();
835 current->total_link_count++;
837 touch_atime(link);
838 nd_set_link(nd, NULL);
840 error = security_inode_follow_link(link->dentry, nd);
841 if (error)
842 goto out_put_nd_path;
844 nd->last_type = LAST_BIND;
845 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
846 error = PTR_ERR(*p);
847 if (IS_ERR(*p))
848 goto out_put_nd_path;
850 error = 0;
851 s = nd_get_link(nd);
852 if (s) {
853 error = __vfs_follow_link(nd, s);
854 if (unlikely(error))
855 put_link(nd, link, *p);
858 return error;
860 out_put_nd_path:
861 *p = NULL;
862 path_put(&nd->path);
863 path_put(link);
864 return error;
867 static int follow_up_rcu(struct path *path)
869 struct mount *mnt = real_mount(path->mnt);
870 struct mount *parent;
871 struct dentry *mountpoint;
873 parent = mnt->mnt_parent;
874 if (&parent->mnt == path->mnt)
875 return 0;
876 mountpoint = mnt->mnt_mountpoint;
877 path->dentry = mountpoint;
878 path->mnt = &parent->mnt;
879 return 1;
883 * follow_up - Find the mountpoint of path's vfsmount
885 * Given a path, find the mountpoint of its source file system.
886 * Replace @path with the path of the mountpoint in the parent mount.
887 * Up is towards /.
889 * Return 1 if we went up a level and 0 if we were already at the
890 * root.
892 int follow_up(struct path *path)
894 struct mount *mnt = real_mount(path->mnt);
895 struct mount *parent;
896 struct dentry *mountpoint;
898 br_read_lock(&vfsmount_lock);
899 parent = mnt->mnt_parent;
900 if (parent == mnt) {
901 br_read_unlock(&vfsmount_lock);
902 return 0;
904 mntget(&parent->mnt);
905 mountpoint = dget(mnt->mnt_mountpoint);
906 br_read_unlock(&vfsmount_lock);
907 dput(path->dentry);
908 path->dentry = mountpoint;
909 mntput(path->mnt);
910 path->mnt = &parent->mnt;
911 return 1;
915 * Perform an automount
916 * - return -EISDIR to tell follow_managed() to stop and return the path we
917 * were called with.
919 static int follow_automount(struct path *path, unsigned flags,
920 bool *need_mntput)
922 struct vfsmount *mnt;
923 int err;
925 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
926 return -EREMOTE;
928 /* We don't want to mount if someone's just doing a stat -
929 * unless they're stat'ing a directory and appended a '/' to
930 * the name.
932 * We do, however, want to mount if someone wants to open or
933 * create a file of any type under the mountpoint, wants to
934 * traverse through the mountpoint or wants to open the
935 * mounted directory. Also, autofs may mark negative dentries
936 * as being automount points. These will need the attentions
937 * of the daemon to instantiate them before they can be used.
939 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
940 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
941 path->dentry->d_inode)
942 return -EISDIR;
944 current->total_link_count++;
945 if (current->total_link_count >= 40)
946 return -ELOOP;
948 mnt = path->dentry->d_op->d_automount(path);
949 if (IS_ERR(mnt)) {
951 * The filesystem is allowed to return -EISDIR here to indicate
952 * it doesn't want to automount. For instance, autofs would do
953 * this so that its userspace daemon can mount on this dentry.
955 * However, we can only permit this if it's a terminal point in
956 * the path being looked up; if it wasn't then the remainder of
957 * the path is inaccessible and we should say so.
959 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
960 return -EREMOTE;
961 return PTR_ERR(mnt);
964 if (!mnt) /* mount collision */
965 return 0;
967 if (!*need_mntput) {
968 /* lock_mount() may release path->mnt on error */
969 mntget(path->mnt);
970 *need_mntput = true;
972 err = finish_automount(mnt, path);
974 switch (err) {
975 case -EBUSY:
976 /* Someone else made a mount here whilst we were busy */
977 return 0;
978 case 0:
979 path_put(path);
980 path->mnt = mnt;
981 path->dentry = dget(mnt->mnt_root);
982 return 0;
983 default:
984 return err;
990 * Handle a dentry that is managed in some way.
991 * - Flagged for transit management (autofs)
992 * - Flagged as mountpoint
993 * - Flagged as automount point
995 * This may only be called in refwalk mode.
997 * Serialization is taken care of in namespace.c
999 static int follow_managed(struct path *path, unsigned flags)
1001 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1002 unsigned managed;
1003 bool need_mntput = false;
1004 int ret = 0;
1006 /* Given that we're not holding a lock here, we retain the value in a
1007 * local variable for each dentry as we look at it so that we don't see
1008 * the components of that value change under us */
1009 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1010 managed &= DCACHE_MANAGED_DENTRY,
1011 unlikely(managed != 0)) {
1012 /* Allow the filesystem to manage the transit without i_mutex
1013 * being held. */
1014 if (managed & DCACHE_MANAGE_TRANSIT) {
1015 BUG_ON(!path->dentry->d_op);
1016 BUG_ON(!path->dentry->d_op->d_manage);
1017 ret = path->dentry->d_op->d_manage(path->dentry, false);
1018 if (ret < 0)
1019 break;
1022 /* Transit to a mounted filesystem. */
1023 if (managed & DCACHE_MOUNTED) {
1024 struct vfsmount *mounted = lookup_mnt(path);
1025 if (mounted) {
1026 dput(path->dentry);
1027 if (need_mntput)
1028 mntput(path->mnt);
1029 path->mnt = mounted;
1030 path->dentry = dget(mounted->mnt_root);
1031 need_mntput = true;
1032 continue;
1035 /* Something is mounted on this dentry in another
1036 * namespace and/or whatever was mounted there in this
1037 * namespace got unmounted before we managed to get the
1038 * vfsmount_lock */
1041 /* Handle an automount point */
1042 if (managed & DCACHE_NEED_AUTOMOUNT) {
1043 ret = follow_automount(path, flags, &need_mntput);
1044 if (ret < 0)
1045 break;
1046 continue;
1049 /* We didn't change the current path point */
1050 break;
1053 if (need_mntput && path->mnt == mnt)
1054 mntput(path->mnt);
1055 if (ret == -EISDIR)
1056 ret = 0;
1057 return ret < 0 ? ret : need_mntput;
1060 int follow_down_one(struct path *path)
1062 struct vfsmount *mounted;
1064 mounted = lookup_mnt(path);
1065 if (mounted) {
1066 dput(path->dentry);
1067 mntput(path->mnt);
1068 path->mnt = mounted;
1069 path->dentry = dget(mounted->mnt_root);
1070 return 1;
1072 return 0;
1075 static inline bool managed_dentry_might_block(struct dentry *dentry)
1077 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1078 dentry->d_op->d_manage(dentry, true) < 0);
1082 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1083 * we meet a managed dentry that would need blocking.
1085 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1086 struct inode **inode)
1088 for (;;) {
1089 struct mount *mounted;
1091 * Don't forget we might have a non-mountpoint managed dentry
1092 * that wants to block transit.
1094 if (unlikely(managed_dentry_might_block(path->dentry)))
1095 return false;
1097 if (!d_mountpoint(path->dentry))
1098 break;
1100 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1101 if (!mounted)
1102 break;
1103 path->mnt = &mounted->mnt;
1104 path->dentry = mounted->mnt.mnt_root;
1105 nd->flags |= LOOKUP_JUMPED;
1106 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1108 * Update the inode too. We don't need to re-check the
1109 * dentry sequence number here after this d_inode read,
1110 * because a mount-point is always pinned.
1112 *inode = path->dentry->d_inode;
1114 return true;
1117 static void follow_mount_rcu(struct nameidata *nd)
1119 while (d_mountpoint(nd->path.dentry)) {
1120 struct mount *mounted;
1121 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
1122 if (!mounted)
1123 break;
1124 nd->path.mnt = &mounted->mnt;
1125 nd->path.dentry = mounted->mnt.mnt_root;
1126 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1130 static int follow_dotdot_rcu(struct nameidata *nd)
1132 set_root_rcu(nd);
1134 while (1) {
1135 if (nd->path.dentry == nd->root.dentry &&
1136 nd->path.mnt == nd->root.mnt) {
1137 break;
1139 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1140 struct dentry *old = nd->path.dentry;
1141 struct dentry *parent = old->d_parent;
1142 unsigned seq;
1144 seq = read_seqcount_begin(&parent->d_seq);
1145 if (read_seqcount_retry(&old->d_seq, nd->seq))
1146 goto failed;
1147 nd->path.dentry = parent;
1148 nd->seq = seq;
1149 break;
1151 if (!follow_up_rcu(&nd->path))
1152 break;
1153 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1155 follow_mount_rcu(nd);
1156 nd->inode = nd->path.dentry->d_inode;
1157 return 0;
1159 failed:
1160 nd->flags &= ~LOOKUP_RCU;
1161 if (!(nd->flags & LOOKUP_ROOT))
1162 nd->root.mnt = NULL;
1163 unlock_rcu_walk();
1164 return -ECHILD;
1168 * Follow down to the covering mount currently visible to userspace. At each
1169 * point, the filesystem owning that dentry may be queried as to whether the
1170 * caller is permitted to proceed or not.
1172 int follow_down(struct path *path)
1174 unsigned managed;
1175 int ret;
1177 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1179 /* Allow the filesystem to manage the transit without i_mutex
1180 * being held.
1182 * We indicate to the filesystem if someone is trying to mount
1183 * something here. This gives autofs the chance to deny anyone
1184 * other than its daemon the right to mount on its
1185 * superstructure.
1187 * The filesystem may sleep at this point.
1189 if (managed & DCACHE_MANAGE_TRANSIT) {
1190 BUG_ON(!path->dentry->d_op);
1191 BUG_ON(!path->dentry->d_op->d_manage);
1192 ret = path->dentry->d_op->d_manage(
1193 path->dentry, false);
1194 if (ret < 0)
1195 return ret == -EISDIR ? 0 : ret;
1198 /* Transit to a mounted filesystem. */
1199 if (managed & DCACHE_MOUNTED) {
1200 struct vfsmount *mounted = lookup_mnt(path);
1201 if (!mounted)
1202 break;
1203 dput(path->dentry);
1204 mntput(path->mnt);
1205 path->mnt = mounted;
1206 path->dentry = dget(mounted->mnt_root);
1207 continue;
1210 /* Don't handle automount points here */
1211 break;
1213 return 0;
1217 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1219 static void follow_mount(struct path *path)
1221 while (d_mountpoint(path->dentry)) {
1222 struct vfsmount *mounted = lookup_mnt(path);
1223 if (!mounted)
1224 break;
1225 dput(path->dentry);
1226 mntput(path->mnt);
1227 path->mnt = mounted;
1228 path->dentry = dget(mounted->mnt_root);
1232 static void follow_dotdot(struct nameidata *nd)
1234 set_root(nd);
1236 while(1) {
1237 struct dentry *old = nd->path.dentry;
1239 if (nd->path.dentry == nd->root.dentry &&
1240 nd->path.mnt == nd->root.mnt) {
1241 break;
1243 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1244 /* rare case of legitimate dget_parent()... */
1245 nd->path.dentry = dget_parent(nd->path.dentry);
1246 dput(old);
1247 break;
1249 if (!follow_up(&nd->path))
1250 break;
1252 follow_mount(&nd->path);
1253 nd->inode = nd->path.dentry->d_inode;
1257 * This looks up the name in dcache, possibly revalidates the old dentry and
1258 * allocates a new one if not found or not valid. In the need_lookup argument
1259 * returns whether i_op->lookup is necessary.
1261 * dir->d_inode->i_mutex must be held
1263 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1264 unsigned int flags, bool *need_lookup)
1266 struct dentry *dentry;
1267 int error;
1269 *need_lookup = false;
1270 dentry = d_lookup(dir, name);
1271 if (dentry) {
1272 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1273 error = d_revalidate(dentry, flags);
1274 if (unlikely(error <= 0)) {
1275 if (error < 0) {
1276 dput(dentry);
1277 return ERR_PTR(error);
1278 } else if (!d_invalidate(dentry)) {
1279 dput(dentry);
1280 dentry = NULL;
1286 if (!dentry) {
1287 dentry = d_alloc(dir, name);
1288 if (unlikely(!dentry))
1289 return ERR_PTR(-ENOMEM);
1291 *need_lookup = true;
1293 return dentry;
1297 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1298 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1300 * dir->d_inode->i_mutex must be held
1302 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1303 unsigned int flags)
1305 struct dentry *old;
1307 /* Don't create child dentry for a dead directory. */
1308 if (unlikely(IS_DEADDIR(dir))) {
1309 dput(dentry);
1310 return ERR_PTR(-ENOENT);
1313 old = dir->i_op->lookup(dir, dentry, flags);
1314 if (unlikely(old)) {
1315 dput(dentry);
1316 dentry = old;
1318 return dentry;
1321 static struct dentry *__lookup_hash(struct qstr *name,
1322 struct dentry *base, unsigned int flags)
1324 bool need_lookup;
1325 struct dentry *dentry;
1327 dentry = lookup_dcache(name, base, flags, &need_lookup);
1328 if (!need_lookup)
1329 return dentry;
1331 return lookup_real(base->d_inode, dentry, flags);
1335 * It's more convoluted than I'd like it to be, but... it's still fairly
1336 * small and for now I'd prefer to have fast path as straight as possible.
1337 * It _is_ time-critical.
1339 static int lookup_fast(struct nameidata *nd,
1340 struct path *path, struct inode **inode)
1342 struct vfsmount *mnt = nd->path.mnt;
1343 struct dentry *dentry, *parent = nd->path.dentry;
1344 int need_reval = 1;
1345 int status = 1;
1346 int err;
1349 * Rename seqlock is not required here because in the off chance
1350 * of a false negative due to a concurrent rename, we're going to
1351 * do the non-racy lookup, below.
1353 if (nd->flags & LOOKUP_RCU) {
1354 unsigned seq;
1355 dentry = __d_lookup_rcu(parent, &nd->last, &seq, nd->inode);
1356 if (!dentry)
1357 goto unlazy;
1360 * This sequence count validates that the inode matches
1361 * the dentry name information from lookup.
1363 *inode = dentry->d_inode;
1364 if (read_seqcount_retry(&dentry->d_seq, seq))
1365 return -ECHILD;
1368 * This sequence count validates that the parent had no
1369 * changes while we did the lookup of the dentry above.
1371 * The memory barrier in read_seqcount_begin of child is
1372 * enough, we can use __read_seqcount_retry here.
1374 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1375 return -ECHILD;
1376 nd->seq = seq;
1378 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1379 status = d_revalidate(dentry, nd->flags);
1380 if (unlikely(status <= 0)) {
1381 if (status != -ECHILD)
1382 need_reval = 0;
1383 goto unlazy;
1386 path->mnt = mnt;
1387 path->dentry = dentry;
1388 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1389 goto unlazy;
1390 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1391 goto unlazy;
1392 return 0;
1393 unlazy:
1394 if (unlazy_walk(nd, dentry))
1395 return -ECHILD;
1396 } else {
1397 dentry = __d_lookup(parent, &nd->last);
1400 if (unlikely(!dentry))
1401 goto need_lookup;
1403 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1404 status = d_revalidate(dentry, nd->flags);
1405 if (unlikely(status <= 0)) {
1406 if (status < 0) {
1407 dput(dentry);
1408 return status;
1410 if (!d_invalidate(dentry)) {
1411 dput(dentry);
1412 goto need_lookup;
1416 path->mnt = mnt;
1417 path->dentry = dentry;
1418 err = follow_managed(path, nd->flags);
1419 if (unlikely(err < 0)) {
1420 path_put_conditional(path, nd);
1421 return err;
1423 if (err)
1424 nd->flags |= LOOKUP_JUMPED;
1425 *inode = path->dentry->d_inode;
1426 return 0;
1428 need_lookup:
1429 return 1;
1432 /* Fast lookup failed, do it the slow way */
1433 static int lookup_slow(struct nameidata *nd, struct path *path)
1435 struct dentry *dentry, *parent;
1436 int err;
1438 parent = nd->path.dentry;
1439 BUG_ON(nd->inode != parent->d_inode);
1441 mutex_lock(&parent->d_inode->i_mutex);
1442 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1443 mutex_unlock(&parent->d_inode->i_mutex);
1444 if (IS_ERR(dentry))
1445 return PTR_ERR(dentry);
1446 path->mnt = nd->path.mnt;
1447 path->dentry = dentry;
1448 err = follow_managed(path, nd->flags);
1449 if (unlikely(err < 0)) {
1450 path_put_conditional(path, nd);
1451 return err;
1453 if (err)
1454 nd->flags |= LOOKUP_JUMPED;
1455 return 0;
1458 static inline int may_lookup(struct nameidata *nd)
1460 if (nd->flags & LOOKUP_RCU) {
1461 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1462 if (err != -ECHILD)
1463 return err;
1464 if (unlazy_walk(nd, NULL))
1465 return -ECHILD;
1467 return inode_permission(nd->inode, MAY_EXEC);
1470 static inline int handle_dots(struct nameidata *nd, int type)
1472 if (type == LAST_DOTDOT) {
1473 if (nd->flags & LOOKUP_RCU) {
1474 if (follow_dotdot_rcu(nd))
1475 return -ECHILD;
1476 } else
1477 follow_dotdot(nd);
1479 return 0;
1482 static void terminate_walk(struct nameidata *nd)
1484 if (!(nd->flags & LOOKUP_RCU)) {
1485 path_put(&nd->path);
1486 } else {
1487 nd->flags &= ~LOOKUP_RCU;
1488 if (!(nd->flags & LOOKUP_ROOT))
1489 nd->root.mnt = NULL;
1490 unlock_rcu_walk();
1495 * Do we need to follow links? We _really_ want to be able
1496 * to do this check without having to look at inode->i_op,
1497 * so we keep a cache of "no, this doesn't need follow_link"
1498 * for the common case.
1500 static inline int should_follow_link(struct inode *inode, int follow)
1502 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1503 if (likely(inode->i_op->follow_link))
1504 return follow;
1506 /* This gets set once for the inode lifetime */
1507 spin_lock(&inode->i_lock);
1508 inode->i_opflags |= IOP_NOFOLLOW;
1509 spin_unlock(&inode->i_lock);
1511 return 0;
1514 static inline int walk_component(struct nameidata *nd, struct path *path,
1515 int follow)
1517 struct inode *inode;
1518 int err;
1520 * "." and ".." are special - ".." especially so because it has
1521 * to be able to know about the current root directory and
1522 * parent relationships.
1524 if (unlikely(nd->last_type != LAST_NORM))
1525 return handle_dots(nd, nd->last_type);
1526 err = lookup_fast(nd, path, &inode);
1527 if (unlikely(err)) {
1528 if (err < 0)
1529 goto out_err;
1531 err = lookup_slow(nd, path);
1532 if (err < 0)
1533 goto out_err;
1535 inode = path->dentry->d_inode;
1537 err = -ENOENT;
1538 if (!inode)
1539 goto out_path_put;
1541 if (should_follow_link(inode, follow)) {
1542 if (nd->flags & LOOKUP_RCU) {
1543 if (unlikely(unlazy_walk(nd, path->dentry))) {
1544 err = -ECHILD;
1545 goto out_err;
1548 BUG_ON(inode != path->dentry->d_inode);
1549 return 1;
1551 path_to_nameidata(path, nd);
1552 nd->inode = inode;
1553 return 0;
1555 out_path_put:
1556 path_to_nameidata(path, nd);
1557 out_err:
1558 terminate_walk(nd);
1559 return err;
1563 * This limits recursive symlink follows to 8, while
1564 * limiting consecutive symlinks to 40.
1566 * Without that kind of total limit, nasty chains of consecutive
1567 * symlinks can cause almost arbitrarily long lookups.
1569 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1571 int res;
1573 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1574 path_put_conditional(path, nd);
1575 path_put(&nd->path);
1576 return -ELOOP;
1578 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1580 nd->depth++;
1581 current->link_count++;
1583 do {
1584 struct path link = *path;
1585 void *cookie;
1587 res = follow_link(&link, nd, &cookie);
1588 if (res)
1589 break;
1590 res = walk_component(nd, path, LOOKUP_FOLLOW);
1591 put_link(nd, &link, cookie);
1592 } while (res > 0);
1594 current->link_count--;
1595 nd->depth--;
1596 return res;
1600 * We really don't want to look at inode->i_op->lookup
1601 * when we don't have to. So we keep a cache bit in
1602 * the inode ->i_opflags field that says "yes, we can
1603 * do lookup on this inode".
1605 static inline int can_lookup(struct inode *inode)
1607 if (likely(inode->i_opflags & IOP_LOOKUP))
1608 return 1;
1609 if (likely(!inode->i_op->lookup))
1610 return 0;
1612 /* We do this once for the lifetime of the inode */
1613 spin_lock(&inode->i_lock);
1614 inode->i_opflags |= IOP_LOOKUP;
1615 spin_unlock(&inode->i_lock);
1616 return 1;
1620 * We can do the critical dentry name comparison and hashing
1621 * operations one word at a time, but we are limited to:
1623 * - Architectures with fast unaligned word accesses. We could
1624 * do a "get_unaligned()" if this helps and is sufficiently
1625 * fast.
1627 * - Little-endian machines (so that we can generate the mask
1628 * of low bytes efficiently). Again, we *could* do a byte
1629 * swapping load on big-endian architectures if that is not
1630 * expensive enough to make the optimization worthless.
1632 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1633 * do not trap on the (extremely unlikely) case of a page
1634 * crossing operation.
1636 * - Furthermore, we need an efficient 64-bit compile for the
1637 * 64-bit case in order to generate the "number of bytes in
1638 * the final mask". Again, that could be replaced with a
1639 * efficient population count instruction or similar.
1641 #ifdef CONFIG_DCACHE_WORD_ACCESS
1643 #include <asm/word-at-a-time.h>
1645 #ifdef CONFIG_64BIT
1647 static inline unsigned int fold_hash(unsigned long hash)
1649 hash += hash >> (8*sizeof(int));
1650 return hash;
1653 #else /* 32-bit case */
1655 #define fold_hash(x) (x)
1657 #endif
1659 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1661 unsigned long a, mask;
1662 unsigned long hash = 0;
1664 for (;;) {
1665 a = load_unaligned_zeropad(name);
1666 if (len < sizeof(unsigned long))
1667 break;
1668 hash += a;
1669 hash *= 9;
1670 name += sizeof(unsigned long);
1671 len -= sizeof(unsigned long);
1672 if (!len)
1673 goto done;
1675 mask = ~(~0ul << len*8);
1676 hash += mask & a;
1677 done:
1678 return fold_hash(hash);
1680 EXPORT_SYMBOL(full_name_hash);
1683 * Calculate the length and hash of the path component, and
1684 * return the length of the component;
1686 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1688 unsigned long a, b, adata, bdata, mask, hash, len;
1689 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1691 hash = a = 0;
1692 len = -sizeof(unsigned long);
1693 do {
1694 hash = (hash + a) * 9;
1695 len += sizeof(unsigned long);
1696 a = load_unaligned_zeropad(name+len);
1697 b = a ^ REPEAT_BYTE('/');
1698 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1700 adata = prep_zero_mask(a, adata, &constants);
1701 bdata = prep_zero_mask(b, bdata, &constants);
1703 mask = create_zero_mask(adata | bdata);
1705 hash += a & zero_bytemask(mask);
1706 *hashp = fold_hash(hash);
1708 return len + find_zero(mask);
1711 #else
1713 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1715 unsigned long hash = init_name_hash();
1716 while (len--)
1717 hash = partial_name_hash(*name++, hash);
1718 return end_name_hash(hash);
1720 EXPORT_SYMBOL(full_name_hash);
1723 * We know there's a real path component here of at least
1724 * one character.
1726 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1728 unsigned long hash = init_name_hash();
1729 unsigned long len = 0, c;
1731 c = (unsigned char)*name;
1732 do {
1733 len++;
1734 hash = partial_name_hash(c, hash);
1735 c = (unsigned char)name[len];
1736 } while (c && c != '/');
1737 *hashp = end_name_hash(hash);
1738 return len;
1741 #endif
1744 * Name resolution.
1745 * This is the basic name resolution function, turning a pathname into
1746 * the final dentry. We expect 'base' to be positive and a directory.
1748 * Returns 0 and nd will have valid dentry and mnt on success.
1749 * Returns error and drops reference to input namei data on failure.
1751 static int link_path_walk(const char *name, struct nameidata *nd)
1753 struct path next;
1754 int err;
1756 while (*name=='/')
1757 name++;
1758 if (!*name)
1759 return 0;
1761 /* At this point we know we have a real path component. */
1762 for(;;) {
1763 struct qstr this;
1764 long len;
1765 int type;
1767 err = may_lookup(nd);
1768 if (err)
1769 break;
1771 len = hash_name(name, &this.hash);
1772 this.name = name;
1773 this.len = len;
1775 type = LAST_NORM;
1776 if (name[0] == '.') switch (len) {
1777 case 2:
1778 if (name[1] == '.') {
1779 type = LAST_DOTDOT;
1780 nd->flags |= LOOKUP_JUMPED;
1782 break;
1783 case 1:
1784 type = LAST_DOT;
1786 if (likely(type == LAST_NORM)) {
1787 struct dentry *parent = nd->path.dentry;
1788 nd->flags &= ~LOOKUP_JUMPED;
1789 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1790 err = parent->d_op->d_hash(parent, nd->inode,
1791 &this);
1792 if (err < 0)
1793 break;
1797 nd->last = this;
1798 nd->last_type = type;
1800 if (!name[len])
1801 return 0;
1803 * If it wasn't NUL, we know it was '/'. Skip that
1804 * slash, and continue until no more slashes.
1806 do {
1807 len++;
1808 } while (unlikely(name[len] == '/'));
1809 if (!name[len])
1810 return 0;
1812 name += len;
1814 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1815 if (err < 0)
1816 return err;
1818 if (err) {
1819 err = nested_symlink(&next, nd);
1820 if (err)
1821 return err;
1823 if (!can_lookup(nd->inode)) {
1824 err = -ENOTDIR;
1825 break;
1828 terminate_walk(nd);
1829 return err;
1832 static int path_init(int dfd, const char *name, unsigned int flags,
1833 struct nameidata *nd, struct file **fp)
1835 int retval = 0;
1837 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1838 nd->flags = flags | LOOKUP_JUMPED;
1839 nd->depth = 0;
1840 if (flags & LOOKUP_ROOT) {
1841 struct inode *inode = nd->root.dentry->d_inode;
1842 if (*name) {
1843 if (!can_lookup(inode))
1844 return -ENOTDIR;
1845 retval = inode_permission(inode, MAY_EXEC);
1846 if (retval)
1847 return retval;
1849 nd->path = nd->root;
1850 nd->inode = inode;
1851 if (flags & LOOKUP_RCU) {
1852 lock_rcu_walk();
1853 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1854 } else {
1855 path_get(&nd->path);
1857 return 0;
1860 nd->root.mnt = NULL;
1862 if (*name=='/') {
1863 if (flags & LOOKUP_RCU) {
1864 lock_rcu_walk();
1865 set_root_rcu(nd);
1866 } else {
1867 set_root(nd);
1868 path_get(&nd->root);
1870 nd->path = nd->root;
1871 } else if (dfd == AT_FDCWD) {
1872 if (flags & LOOKUP_RCU) {
1873 struct fs_struct *fs = current->fs;
1874 unsigned seq;
1876 lock_rcu_walk();
1878 do {
1879 seq = read_seqcount_begin(&fs->seq);
1880 nd->path = fs->pwd;
1881 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1882 } while (read_seqcount_retry(&fs->seq, seq));
1883 } else {
1884 get_fs_pwd(current->fs, &nd->path);
1886 } else {
1887 /* Caller must check execute permissions on the starting path component */
1888 struct fd f = fdget_raw(dfd);
1889 struct dentry *dentry;
1891 if (!f.file)
1892 return -EBADF;
1894 dentry = f.file->f_path.dentry;
1896 if (*name) {
1897 if (!can_lookup(dentry->d_inode)) {
1898 fdput(f);
1899 return -ENOTDIR;
1903 nd->path = f.file->f_path;
1904 if (flags & LOOKUP_RCU) {
1905 if (f.need_put)
1906 *fp = f.file;
1907 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1908 lock_rcu_walk();
1909 } else {
1910 path_get(&nd->path);
1911 fdput(f);
1915 nd->inode = nd->path.dentry->d_inode;
1916 return 0;
1919 static inline int lookup_last(struct nameidata *nd, struct path *path)
1921 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1922 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1924 nd->flags &= ~LOOKUP_PARENT;
1925 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1928 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1929 static int path_lookupat(int dfd, const char *name,
1930 unsigned int flags, struct nameidata *nd)
1932 struct file *base = NULL;
1933 struct path path;
1934 int err;
1937 * Path walking is largely split up into 2 different synchronisation
1938 * schemes, rcu-walk and ref-walk (explained in
1939 * Documentation/filesystems/path-lookup.txt). These share much of the
1940 * path walk code, but some things particularly setup, cleanup, and
1941 * following mounts are sufficiently divergent that functions are
1942 * duplicated. Typically there is a function foo(), and its RCU
1943 * analogue, foo_rcu().
1945 * -ECHILD is the error number of choice (just to avoid clashes) that
1946 * is returned if some aspect of an rcu-walk fails. Such an error must
1947 * be handled by restarting a traditional ref-walk (which will always
1948 * be able to complete).
1950 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1952 if (unlikely(err))
1953 return err;
1955 current->total_link_count = 0;
1956 err = link_path_walk(name, nd);
1958 if (!err && !(flags & LOOKUP_PARENT)) {
1959 err = lookup_last(nd, &path);
1960 while (err > 0) {
1961 void *cookie;
1962 struct path link = path;
1963 err = may_follow_link(&link, nd);
1964 if (unlikely(err))
1965 break;
1966 nd->flags |= LOOKUP_PARENT;
1967 err = follow_link(&link, nd, &cookie);
1968 if (err)
1969 break;
1970 err = lookup_last(nd, &path);
1971 put_link(nd, &link, cookie);
1975 if (!err)
1976 err = complete_walk(nd);
1978 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1979 if (!nd->inode->i_op->lookup) {
1980 path_put(&nd->path);
1981 err = -ENOTDIR;
1985 if (base)
1986 fput(base);
1988 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1989 path_put(&nd->root);
1990 nd->root.mnt = NULL;
1992 return err;
1995 static int filename_lookup(int dfd, struct filename *name,
1996 unsigned int flags, struct nameidata *nd)
1998 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1999 if (unlikely(retval == -ECHILD))
2000 retval = path_lookupat(dfd, name->name, flags, nd);
2001 if (unlikely(retval == -ESTALE))
2002 retval = path_lookupat(dfd, name->name,
2003 flags | LOOKUP_REVAL, nd);
2005 if (likely(!retval))
2006 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2007 return retval;
2010 static int do_path_lookup(int dfd, const char *name,
2011 unsigned int flags, struct nameidata *nd)
2013 struct filename filename = { .name = name };
2015 return filename_lookup(dfd, &filename, flags, nd);
2018 /* does lookup, returns the object with parent locked */
2019 struct dentry *kern_path_locked(const char *name, struct path *path)
2021 struct nameidata nd;
2022 struct dentry *d;
2023 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2024 if (err)
2025 return ERR_PTR(err);
2026 if (nd.last_type != LAST_NORM) {
2027 path_put(&nd.path);
2028 return ERR_PTR(-EINVAL);
2030 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2031 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2032 if (IS_ERR(d)) {
2033 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2034 path_put(&nd.path);
2035 return d;
2037 *path = nd.path;
2038 return d;
2041 int kern_path(const char *name, unsigned int flags, struct path *path)
2043 struct nameidata nd;
2044 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2045 if (!res)
2046 *path = nd.path;
2047 return res;
2051 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2052 * @dentry: pointer to dentry of the base directory
2053 * @mnt: pointer to vfs mount of the base directory
2054 * @name: pointer to file name
2055 * @flags: lookup flags
2056 * @path: pointer to struct path to fill
2058 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2059 const char *name, unsigned int flags,
2060 struct path *path)
2062 struct nameidata nd;
2063 int err;
2064 nd.root.dentry = dentry;
2065 nd.root.mnt = mnt;
2066 BUG_ON(flags & LOOKUP_PARENT);
2067 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2068 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2069 if (!err)
2070 *path = nd.path;
2071 return err;
2075 * Restricted form of lookup. Doesn't follow links, single-component only,
2076 * needs parent already locked. Doesn't follow mounts.
2077 * SMP-safe.
2079 static struct dentry *lookup_hash(struct nameidata *nd)
2081 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2085 * lookup_one_len - filesystem helper to lookup single pathname component
2086 * @name: pathname component to lookup
2087 * @base: base directory to lookup from
2088 * @len: maximum length @len should be interpreted to
2090 * Note that this routine is purely a helper for filesystem usage and should
2091 * not be called by generic code. Also note that by using this function the
2092 * nameidata argument is passed to the filesystem methods and a filesystem
2093 * using this helper needs to be prepared for that.
2095 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2097 struct qstr this;
2098 unsigned int c;
2099 int err;
2101 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2103 this.name = name;
2104 this.len = len;
2105 this.hash = full_name_hash(name, len);
2106 if (!len)
2107 return ERR_PTR(-EACCES);
2109 if (unlikely(name[0] == '.')) {
2110 if (len < 2 || (len == 2 && name[1] == '.'))
2111 return ERR_PTR(-EACCES);
2114 while (len--) {
2115 c = *(const unsigned char *)name++;
2116 if (c == '/' || c == '\0')
2117 return ERR_PTR(-EACCES);
2120 * See if the low-level filesystem might want
2121 * to use its own hash..
2123 if (base->d_flags & DCACHE_OP_HASH) {
2124 int err = base->d_op->d_hash(base, base->d_inode, &this);
2125 if (err < 0)
2126 return ERR_PTR(err);
2129 err = inode_permission(base->d_inode, MAY_EXEC);
2130 if (err)
2131 return ERR_PTR(err);
2133 return __lookup_hash(&this, base, 0);
2136 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2137 struct path *path, int *empty)
2139 struct nameidata nd;
2140 struct filename *tmp = getname_flags(name, flags, empty);
2141 int err = PTR_ERR(tmp);
2142 if (!IS_ERR(tmp)) {
2144 BUG_ON(flags & LOOKUP_PARENT);
2146 err = filename_lookup(dfd, tmp, flags, &nd);
2147 putname(tmp);
2148 if (!err)
2149 *path = nd.path;
2151 return err;
2154 int user_path_at(int dfd, const char __user *name, unsigned flags,
2155 struct path *path)
2157 return user_path_at_empty(dfd, name, flags, path, NULL);
2161 * NB: most callers don't do anything directly with the reference to the
2162 * to struct filename, but the nd->last pointer points into the name string
2163 * allocated by getname. So we must hold the reference to it until all
2164 * path-walking is complete.
2166 static struct filename *
2167 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2168 unsigned int flags)
2170 struct filename *s = getname(path);
2171 int error;
2173 /* only LOOKUP_REVAL is allowed in extra flags */
2174 flags &= LOOKUP_REVAL;
2176 if (IS_ERR(s))
2177 return s;
2179 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2180 if (error) {
2181 putname(s);
2182 return ERR_PTR(error);
2185 return s;
2189 * It's inline, so penalty for filesystems that don't use sticky bit is
2190 * minimal.
2192 static inline int check_sticky(struct inode *dir, struct inode *inode)
2194 kuid_t fsuid = current_fsuid();
2196 if (!(dir->i_mode & S_ISVTX))
2197 return 0;
2198 if (uid_eq(inode->i_uid, fsuid))
2199 return 0;
2200 if (uid_eq(dir->i_uid, fsuid))
2201 return 0;
2202 return !inode_capable(inode, CAP_FOWNER);
2206 * Check whether we can remove a link victim from directory dir, check
2207 * whether the type of victim is right.
2208 * 1. We can't do it if dir is read-only (done in permission())
2209 * 2. We should have write and exec permissions on dir
2210 * 3. We can't remove anything from append-only dir
2211 * 4. We can't do anything with immutable dir (done in permission())
2212 * 5. If the sticky bit on dir is set we should either
2213 * a. be owner of dir, or
2214 * b. be owner of victim, or
2215 * c. have CAP_FOWNER capability
2216 * 6. If the victim is append-only or immutable we can't do antyhing with
2217 * links pointing to it.
2218 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2219 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2220 * 9. We can't remove a root or mountpoint.
2221 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2222 * nfs_async_unlink().
2224 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2226 int error;
2228 if (!victim->d_inode)
2229 return -ENOENT;
2231 BUG_ON(victim->d_parent->d_inode != dir);
2232 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2234 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2235 if (error)
2236 return error;
2237 if (IS_APPEND(dir))
2238 return -EPERM;
2239 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2240 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2241 return -EPERM;
2242 if (isdir) {
2243 if (!S_ISDIR(victim->d_inode->i_mode))
2244 return -ENOTDIR;
2245 if (IS_ROOT(victim))
2246 return -EBUSY;
2247 } else if (S_ISDIR(victim->d_inode->i_mode))
2248 return -EISDIR;
2249 if (IS_DEADDIR(dir))
2250 return -ENOENT;
2251 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2252 return -EBUSY;
2253 return 0;
2256 /* Check whether we can create an object with dentry child in directory
2257 * dir.
2258 * 1. We can't do it if child already exists (open has special treatment for
2259 * this case, but since we are inlined it's OK)
2260 * 2. We can't do it if dir is read-only (done in permission())
2261 * 3. We should have write and exec permissions on dir
2262 * 4. We can't do it if dir is immutable (done in permission())
2264 static inline int may_create(struct inode *dir, struct dentry *child)
2266 if (child->d_inode)
2267 return -EEXIST;
2268 if (IS_DEADDIR(dir))
2269 return -ENOENT;
2270 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2274 * p1 and p2 should be directories on the same fs.
2276 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2278 struct dentry *p;
2280 if (p1 == p2) {
2281 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2282 return NULL;
2285 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2287 p = d_ancestor(p2, p1);
2288 if (p) {
2289 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2290 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2291 return p;
2294 p = d_ancestor(p1, p2);
2295 if (p) {
2296 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2297 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2298 return p;
2301 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2302 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2303 return NULL;
2306 void unlock_rename(struct dentry *p1, struct dentry *p2)
2308 mutex_unlock(&p1->d_inode->i_mutex);
2309 if (p1 != p2) {
2310 mutex_unlock(&p2->d_inode->i_mutex);
2311 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2315 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2316 bool want_excl)
2318 int error = may_create(dir, dentry);
2319 if (error)
2320 return error;
2322 if (!dir->i_op->create)
2323 return -EACCES; /* shouldn't it be ENOSYS? */
2324 mode &= S_IALLUGO;
2325 mode |= S_IFREG;
2326 error = security_inode_create(dir, dentry, mode);
2327 if (error)
2328 return error;
2329 error = dir->i_op->create(dir, dentry, mode, want_excl);
2330 if (!error)
2331 fsnotify_create(dir, dentry);
2332 return error;
2335 static int may_open(struct path *path, int acc_mode, int flag)
2337 struct dentry *dentry = path->dentry;
2338 struct inode *inode = dentry->d_inode;
2339 int error;
2341 /* O_PATH? */
2342 if (!acc_mode)
2343 return 0;
2345 if (!inode)
2346 return -ENOENT;
2348 switch (inode->i_mode & S_IFMT) {
2349 case S_IFLNK:
2350 return -ELOOP;
2351 case S_IFDIR:
2352 if (acc_mode & MAY_WRITE)
2353 return -EISDIR;
2354 break;
2355 case S_IFBLK:
2356 case S_IFCHR:
2357 if (path->mnt->mnt_flags & MNT_NODEV)
2358 return -EACCES;
2359 /*FALLTHRU*/
2360 case S_IFIFO:
2361 case S_IFSOCK:
2362 flag &= ~O_TRUNC;
2363 break;
2366 error = inode_permission(inode, acc_mode);
2367 if (error)
2368 return error;
2371 * An append-only file must be opened in append mode for writing.
2373 if (IS_APPEND(inode)) {
2374 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2375 return -EPERM;
2376 if (flag & O_TRUNC)
2377 return -EPERM;
2380 /* O_NOATIME can only be set by the owner or superuser */
2381 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2382 return -EPERM;
2384 return 0;
2387 static int handle_truncate(struct file *filp)
2389 struct path *path = &filp->f_path;
2390 struct inode *inode = path->dentry->d_inode;
2391 int error = get_write_access(inode);
2392 if (error)
2393 return error;
2395 * Refuse to truncate files with mandatory locks held on them.
2397 error = locks_verify_locked(inode);
2398 if (!error)
2399 error = security_path_truncate(path);
2400 if (!error) {
2401 error = do_truncate(path->dentry, 0,
2402 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2403 filp);
2405 put_write_access(inode);
2406 return error;
2409 static inline int open_to_namei_flags(int flag)
2411 if ((flag & O_ACCMODE) == 3)
2412 flag--;
2413 return flag;
2416 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2418 int error = security_path_mknod(dir, dentry, mode, 0);
2419 if (error)
2420 return error;
2422 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2423 if (error)
2424 return error;
2426 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2430 * Attempt to atomically look up, create and open a file from a negative
2431 * dentry.
2433 * Returns 0 if successful. The file will have been created and attached to
2434 * @file by the filesystem calling finish_open().
2436 * Returns 1 if the file was looked up only or didn't need creating. The
2437 * caller will need to perform the open themselves. @path will have been
2438 * updated to point to the new dentry. This may be negative.
2440 * Returns an error code otherwise.
2442 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2443 struct path *path, struct file *file,
2444 const struct open_flags *op,
2445 bool got_write, bool need_lookup,
2446 int *opened)
2448 struct inode *dir = nd->path.dentry->d_inode;
2449 unsigned open_flag = open_to_namei_flags(op->open_flag);
2450 umode_t mode;
2451 int error;
2452 int acc_mode;
2453 int create_error = 0;
2454 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2456 BUG_ON(dentry->d_inode);
2458 /* Don't create child dentry for a dead directory. */
2459 if (unlikely(IS_DEADDIR(dir))) {
2460 error = -ENOENT;
2461 goto out;
2464 mode = op->mode;
2465 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2466 mode &= ~current_umask();
2468 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
2469 open_flag &= ~O_TRUNC;
2470 *opened |= FILE_CREATED;
2474 * Checking write permission is tricky, bacuse we don't know if we are
2475 * going to actually need it: O_CREAT opens should work as long as the
2476 * file exists. But checking existence breaks atomicity. The trick is
2477 * to check access and if not granted clear O_CREAT from the flags.
2479 * Another problem is returing the "right" error value (e.g. for an
2480 * O_EXCL open we want to return EEXIST not EROFS).
2482 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2483 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2484 if (!(open_flag & O_CREAT)) {
2486 * No O_CREATE -> atomicity not a requirement -> fall
2487 * back to lookup + open
2489 goto no_open;
2490 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2491 /* Fall back and fail with the right error */
2492 create_error = -EROFS;
2493 goto no_open;
2494 } else {
2495 /* No side effects, safe to clear O_CREAT */
2496 create_error = -EROFS;
2497 open_flag &= ~O_CREAT;
2501 if (open_flag & O_CREAT) {
2502 error = may_o_create(&nd->path, dentry, mode);
2503 if (error) {
2504 create_error = error;
2505 if (open_flag & O_EXCL)
2506 goto no_open;
2507 open_flag &= ~O_CREAT;
2511 if (nd->flags & LOOKUP_DIRECTORY)
2512 open_flag |= O_DIRECTORY;
2514 file->f_path.dentry = DENTRY_NOT_SET;
2515 file->f_path.mnt = nd->path.mnt;
2516 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2517 opened);
2518 if (error < 0) {
2519 if (create_error && error == -ENOENT)
2520 error = create_error;
2521 goto out;
2524 acc_mode = op->acc_mode;
2525 if (*opened & FILE_CREATED) {
2526 fsnotify_create(dir, dentry);
2527 acc_mode = MAY_OPEN;
2530 if (error) { /* returned 1, that is */
2531 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2532 error = -EIO;
2533 goto out;
2535 if (file->f_path.dentry) {
2536 dput(dentry);
2537 dentry = file->f_path.dentry;
2539 if (create_error && dentry->d_inode == NULL) {
2540 error = create_error;
2541 goto out;
2543 goto looked_up;
2547 * We didn't have the inode before the open, so check open permission
2548 * here.
2550 error = may_open(&file->f_path, acc_mode, open_flag);
2551 if (error)
2552 fput(file);
2554 out:
2555 dput(dentry);
2556 return error;
2558 no_open:
2559 if (need_lookup) {
2560 dentry = lookup_real(dir, dentry, nd->flags);
2561 if (IS_ERR(dentry))
2562 return PTR_ERR(dentry);
2564 if (create_error) {
2565 int open_flag = op->open_flag;
2567 error = create_error;
2568 if ((open_flag & O_EXCL)) {
2569 if (!dentry->d_inode)
2570 goto out;
2571 } else if (!dentry->d_inode) {
2572 goto out;
2573 } else if ((open_flag & O_TRUNC) &&
2574 S_ISREG(dentry->d_inode->i_mode)) {
2575 goto out;
2577 /* will fail later, go on to get the right error */
2580 looked_up:
2581 path->dentry = dentry;
2582 path->mnt = nd->path.mnt;
2583 return 1;
2587 * Look up and maybe create and open the last component.
2589 * Must be called with i_mutex held on parent.
2591 * Returns 0 if the file was successfully atomically created (if necessary) and
2592 * opened. In this case the file will be returned attached to @file.
2594 * Returns 1 if the file was not completely opened at this time, though lookups
2595 * and creations will have been performed and the dentry returned in @path will
2596 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2597 * specified then a negative dentry may be returned.
2599 * An error code is returned otherwise.
2601 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2602 * cleared otherwise prior to returning.
2604 static int lookup_open(struct nameidata *nd, struct path *path,
2605 struct file *file,
2606 const struct open_flags *op,
2607 bool got_write, int *opened)
2609 struct dentry *dir = nd->path.dentry;
2610 struct inode *dir_inode = dir->d_inode;
2611 struct dentry *dentry;
2612 int error;
2613 bool need_lookup;
2615 *opened &= ~FILE_CREATED;
2616 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2617 if (IS_ERR(dentry))
2618 return PTR_ERR(dentry);
2620 /* Cached positive dentry: will open in f_op->open */
2621 if (!need_lookup && dentry->d_inode)
2622 goto out_no_open;
2624 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2625 return atomic_open(nd, dentry, path, file, op, got_write,
2626 need_lookup, opened);
2629 if (need_lookup) {
2630 BUG_ON(dentry->d_inode);
2632 dentry = lookup_real(dir_inode, dentry, nd->flags);
2633 if (IS_ERR(dentry))
2634 return PTR_ERR(dentry);
2637 /* Negative dentry, just create the file */
2638 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2639 umode_t mode = op->mode;
2640 if (!IS_POSIXACL(dir->d_inode))
2641 mode &= ~current_umask();
2643 * This write is needed to ensure that a
2644 * rw->ro transition does not occur between
2645 * the time when the file is created and when
2646 * a permanent write count is taken through
2647 * the 'struct file' in finish_open().
2649 if (!got_write) {
2650 error = -EROFS;
2651 goto out_dput;
2653 *opened |= FILE_CREATED;
2654 error = security_path_mknod(&nd->path, dentry, mode, 0);
2655 if (error)
2656 goto out_dput;
2657 error = vfs_create(dir->d_inode, dentry, mode,
2658 nd->flags & LOOKUP_EXCL);
2659 if (error)
2660 goto out_dput;
2662 out_no_open:
2663 path->dentry = dentry;
2664 path->mnt = nd->path.mnt;
2665 return 1;
2667 out_dput:
2668 dput(dentry);
2669 return error;
2673 * Handle the last step of open()
2675 static int do_last(struct nameidata *nd, struct path *path,
2676 struct file *file, const struct open_flags *op,
2677 int *opened, struct filename *name)
2679 struct dentry *dir = nd->path.dentry;
2680 int open_flag = op->open_flag;
2681 bool will_truncate = (open_flag & O_TRUNC) != 0;
2682 bool got_write = false;
2683 int acc_mode = op->acc_mode;
2684 struct inode *inode;
2685 bool symlink_ok = false;
2686 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2687 bool retried = false;
2688 int error;
2690 nd->flags &= ~LOOKUP_PARENT;
2691 nd->flags |= op->intent;
2693 switch (nd->last_type) {
2694 case LAST_DOTDOT:
2695 case LAST_DOT:
2696 error = handle_dots(nd, nd->last_type);
2697 if (error)
2698 return error;
2699 /* fallthrough */
2700 case LAST_ROOT:
2701 error = complete_walk(nd);
2702 if (error)
2703 return error;
2704 audit_inode(name, nd->path.dentry, 0);
2705 if (open_flag & O_CREAT) {
2706 error = -EISDIR;
2707 goto out;
2709 goto finish_open;
2710 case LAST_BIND:
2711 error = complete_walk(nd);
2712 if (error)
2713 return error;
2714 audit_inode(name, dir, 0);
2715 goto finish_open;
2718 if (!(open_flag & O_CREAT)) {
2719 if (nd->last.name[nd->last.len])
2720 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2721 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2722 symlink_ok = true;
2723 /* we _can_ be in RCU mode here */
2724 error = lookup_fast(nd, path, &inode);
2725 if (likely(!error))
2726 goto finish_lookup;
2728 if (error < 0)
2729 goto out;
2731 BUG_ON(nd->inode != dir->d_inode);
2732 } else {
2733 /* create side of things */
2735 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2736 * has been cleared when we got to the last component we are
2737 * about to look up
2739 error = complete_walk(nd);
2740 if (error)
2741 return error;
2743 audit_inode(name, dir, 0);
2744 error = -EISDIR;
2745 /* trailing slashes? */
2746 if (nd->last.name[nd->last.len])
2747 goto out;
2750 retry_lookup:
2751 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2752 error = mnt_want_write(nd->path.mnt);
2753 if (!error)
2754 got_write = true;
2756 * do _not_ fail yet - we might not need that or fail with
2757 * a different error; let lookup_open() decide; we'll be
2758 * dropping this one anyway.
2761 mutex_lock(&dir->d_inode->i_mutex);
2762 error = lookup_open(nd, path, file, op, got_write, opened);
2763 mutex_unlock(&dir->d_inode->i_mutex);
2765 if (error <= 0) {
2766 if (error)
2767 goto out;
2769 if ((*opened & FILE_CREATED) ||
2770 !S_ISREG(file_inode(file)->i_mode))
2771 will_truncate = false;
2773 audit_inode(name, file->f_path.dentry, 0);
2774 goto opened;
2777 if (*opened & FILE_CREATED) {
2778 /* Don't check for write permission, don't truncate */
2779 open_flag &= ~O_TRUNC;
2780 will_truncate = false;
2781 acc_mode = MAY_OPEN;
2782 path_to_nameidata(path, nd);
2783 goto finish_open_created;
2787 * create/update audit record if it already exists.
2789 if (path->dentry->d_inode)
2790 audit_inode(name, path->dentry, 0);
2793 * If atomic_open() acquired write access it is dropped now due to
2794 * possible mount and symlink following (this might be optimized away if
2795 * necessary...)
2797 if (got_write) {
2798 mnt_drop_write(nd->path.mnt);
2799 got_write = false;
2802 error = -EEXIST;
2803 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2804 goto exit_dput;
2806 error = follow_managed(path, nd->flags);
2807 if (error < 0)
2808 goto exit_dput;
2810 if (error)
2811 nd->flags |= LOOKUP_JUMPED;
2813 BUG_ON(nd->flags & LOOKUP_RCU);
2814 inode = path->dentry->d_inode;
2815 finish_lookup:
2816 /* we _can_ be in RCU mode here */
2817 error = -ENOENT;
2818 if (!inode) {
2819 path_to_nameidata(path, nd);
2820 goto out;
2823 if (should_follow_link(inode, !symlink_ok)) {
2824 if (nd->flags & LOOKUP_RCU) {
2825 if (unlikely(unlazy_walk(nd, path->dentry))) {
2826 error = -ECHILD;
2827 goto out;
2830 BUG_ON(inode != path->dentry->d_inode);
2831 return 1;
2834 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2835 path_to_nameidata(path, nd);
2836 } else {
2837 save_parent.dentry = nd->path.dentry;
2838 save_parent.mnt = mntget(path->mnt);
2839 nd->path.dentry = path->dentry;
2842 nd->inode = inode;
2843 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2844 error = complete_walk(nd);
2845 if (error) {
2846 path_put(&save_parent);
2847 return error;
2849 error = -EISDIR;
2850 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2851 goto out;
2852 error = -ENOTDIR;
2853 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2854 goto out;
2855 audit_inode(name, nd->path.dentry, 0);
2856 finish_open:
2857 if (!S_ISREG(nd->inode->i_mode))
2858 will_truncate = false;
2860 if (will_truncate) {
2861 error = mnt_want_write(nd->path.mnt);
2862 if (error)
2863 goto out;
2864 got_write = true;
2866 finish_open_created:
2867 error = may_open(&nd->path, acc_mode, open_flag);
2868 if (error)
2869 goto out;
2870 file->f_path.mnt = nd->path.mnt;
2871 error = finish_open(file, nd->path.dentry, NULL, opened);
2872 if (error) {
2873 if (error == -EOPENSTALE)
2874 goto stale_open;
2875 goto out;
2877 opened:
2878 error = open_check_o_direct(file);
2879 if (error)
2880 goto exit_fput;
2881 error = ima_file_check(file, op->acc_mode);
2882 if (error)
2883 goto exit_fput;
2885 if (will_truncate) {
2886 error = handle_truncate(file);
2887 if (error)
2888 goto exit_fput;
2890 out:
2891 if (got_write)
2892 mnt_drop_write(nd->path.mnt);
2893 path_put(&save_parent);
2894 terminate_walk(nd);
2895 return error;
2897 exit_dput:
2898 path_put_conditional(path, nd);
2899 goto out;
2900 exit_fput:
2901 fput(file);
2902 goto out;
2904 stale_open:
2905 /* If no saved parent or already retried then can't retry */
2906 if (!save_parent.dentry || retried)
2907 goto out;
2909 BUG_ON(save_parent.dentry != dir);
2910 path_put(&nd->path);
2911 nd->path = save_parent;
2912 nd->inode = dir->d_inode;
2913 save_parent.mnt = NULL;
2914 save_parent.dentry = NULL;
2915 if (got_write) {
2916 mnt_drop_write(nd->path.mnt);
2917 got_write = false;
2919 retried = true;
2920 goto retry_lookup;
2923 static struct file *path_openat(int dfd, struct filename *pathname,
2924 struct nameidata *nd, const struct open_flags *op, int flags)
2926 struct file *base = NULL;
2927 struct file *file;
2928 struct path path;
2929 int opened = 0;
2930 int error;
2932 file = get_empty_filp();
2933 if (IS_ERR(file))
2934 return file;
2936 file->f_flags = op->open_flag;
2938 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
2939 if (unlikely(error))
2940 goto out;
2942 current->total_link_count = 0;
2943 error = link_path_walk(pathname->name, nd);
2944 if (unlikely(error))
2945 goto out;
2947 error = do_last(nd, &path, file, op, &opened, pathname);
2948 while (unlikely(error > 0)) { /* trailing symlink */
2949 struct path link = path;
2950 void *cookie;
2951 if (!(nd->flags & LOOKUP_FOLLOW)) {
2952 path_put_conditional(&path, nd);
2953 path_put(&nd->path);
2954 error = -ELOOP;
2955 break;
2957 error = may_follow_link(&link, nd);
2958 if (unlikely(error))
2959 break;
2960 nd->flags |= LOOKUP_PARENT;
2961 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2962 error = follow_link(&link, nd, &cookie);
2963 if (unlikely(error))
2964 break;
2965 error = do_last(nd, &path, file, op, &opened, pathname);
2966 put_link(nd, &link, cookie);
2968 out:
2969 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2970 path_put(&nd->root);
2971 if (base)
2972 fput(base);
2973 if (!(opened & FILE_OPENED)) {
2974 BUG_ON(!error);
2975 put_filp(file);
2977 if (unlikely(error)) {
2978 if (error == -EOPENSTALE) {
2979 if (flags & LOOKUP_RCU)
2980 error = -ECHILD;
2981 else
2982 error = -ESTALE;
2984 file = ERR_PTR(error);
2986 return file;
2989 struct file *do_filp_open(int dfd, struct filename *pathname,
2990 const struct open_flags *op, int flags)
2992 struct nameidata nd;
2993 struct file *filp;
2995 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2996 if (unlikely(filp == ERR_PTR(-ECHILD)))
2997 filp = path_openat(dfd, pathname, &nd, op, flags);
2998 if (unlikely(filp == ERR_PTR(-ESTALE)))
2999 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3000 return filp;
3003 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3004 const char *name, const struct open_flags *op, int flags)
3006 struct nameidata nd;
3007 struct file *file;
3008 struct filename filename = { .name = name };
3010 nd.root.mnt = mnt;
3011 nd.root.dentry = dentry;
3013 flags |= LOOKUP_ROOT;
3015 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
3016 return ERR_PTR(-ELOOP);
3018 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3019 if (unlikely(file == ERR_PTR(-ECHILD)))
3020 file = path_openat(-1, &filename, &nd, op, flags);
3021 if (unlikely(file == ERR_PTR(-ESTALE)))
3022 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3023 return file;
3026 struct dentry *kern_path_create(int dfd, const char *pathname,
3027 struct path *path, unsigned int lookup_flags)
3029 struct dentry *dentry = ERR_PTR(-EEXIST);
3030 struct nameidata nd;
3031 int err2;
3032 int error;
3033 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3036 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3037 * other flags passed in are ignored!
3039 lookup_flags &= LOOKUP_REVAL;
3041 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3042 if (error)
3043 return ERR_PTR(error);
3046 * Yucky last component or no last component at all?
3047 * (foo/., foo/.., /////)
3049 if (nd.last_type != LAST_NORM)
3050 goto out;
3051 nd.flags &= ~LOOKUP_PARENT;
3052 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3054 /* don't fail immediately if it's r/o, at least try to report other errors */
3055 err2 = mnt_want_write(nd.path.mnt);
3057 * Do the final lookup.
3059 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3060 dentry = lookup_hash(&nd);
3061 if (IS_ERR(dentry))
3062 goto unlock;
3064 error = -EEXIST;
3065 if (dentry->d_inode)
3066 goto fail;
3068 * Special case - lookup gave negative, but... we had foo/bar/
3069 * From the vfs_mknod() POV we just have a negative dentry -
3070 * all is fine. Let's be bastards - you had / on the end, you've
3071 * been asking for (non-existent) directory. -ENOENT for you.
3073 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3074 error = -ENOENT;
3075 goto fail;
3077 if (unlikely(err2)) {
3078 error = err2;
3079 goto fail;
3081 *path = nd.path;
3082 return dentry;
3083 fail:
3084 dput(dentry);
3085 dentry = ERR_PTR(error);
3086 unlock:
3087 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3088 if (!err2)
3089 mnt_drop_write(nd.path.mnt);
3090 out:
3091 path_put(&nd.path);
3092 return dentry;
3094 EXPORT_SYMBOL(kern_path_create);
3096 void done_path_create(struct path *path, struct dentry *dentry)
3098 dput(dentry);
3099 mutex_unlock(&path->dentry->d_inode->i_mutex);
3100 mnt_drop_write(path->mnt);
3101 path_put(path);
3103 EXPORT_SYMBOL(done_path_create);
3105 struct dentry *user_path_create(int dfd, const char __user *pathname,
3106 struct path *path, unsigned int lookup_flags)
3108 struct filename *tmp = getname(pathname);
3109 struct dentry *res;
3110 if (IS_ERR(tmp))
3111 return ERR_CAST(tmp);
3112 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3113 putname(tmp);
3114 return res;
3116 EXPORT_SYMBOL(user_path_create);
3118 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3120 int error = may_create(dir, dentry);
3122 if (error)
3123 return error;
3125 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3126 return -EPERM;
3128 if (!dir->i_op->mknod)
3129 return -EPERM;
3131 error = devcgroup_inode_mknod(mode, dev);
3132 if (error)
3133 return error;
3135 error = security_inode_mknod(dir, dentry, mode, dev);
3136 if (error)
3137 return error;
3139 error = dir->i_op->mknod(dir, dentry, mode, dev);
3140 if (!error)
3141 fsnotify_create(dir, dentry);
3142 return error;
3145 static int may_mknod(umode_t mode)
3147 switch (mode & S_IFMT) {
3148 case S_IFREG:
3149 case S_IFCHR:
3150 case S_IFBLK:
3151 case S_IFIFO:
3152 case S_IFSOCK:
3153 case 0: /* zero mode translates to S_IFREG */
3154 return 0;
3155 case S_IFDIR:
3156 return -EPERM;
3157 default:
3158 return -EINVAL;
3162 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3163 unsigned, dev)
3165 struct dentry *dentry;
3166 struct path path;
3167 int error;
3168 unsigned int lookup_flags = 0;
3170 error = may_mknod(mode);
3171 if (error)
3172 return error;
3173 retry:
3174 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3175 if (IS_ERR(dentry))
3176 return PTR_ERR(dentry);
3178 if (!IS_POSIXACL(path.dentry->d_inode))
3179 mode &= ~current_umask();
3180 error = security_path_mknod(&path, dentry, mode, dev);
3181 if (error)
3182 goto out;
3183 switch (mode & S_IFMT) {
3184 case 0: case S_IFREG:
3185 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3186 break;
3187 case S_IFCHR: case S_IFBLK:
3188 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3189 new_decode_dev(dev));
3190 break;
3191 case S_IFIFO: case S_IFSOCK:
3192 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3193 break;
3195 out:
3196 done_path_create(&path, dentry);
3197 if (retry_estale(error, lookup_flags)) {
3198 lookup_flags |= LOOKUP_REVAL;
3199 goto retry;
3201 return error;
3204 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3206 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3209 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3211 int error = may_create(dir, dentry);
3212 unsigned max_links = dir->i_sb->s_max_links;
3214 if (error)
3215 return error;
3217 if (!dir->i_op->mkdir)
3218 return -EPERM;
3220 mode &= (S_IRWXUGO|S_ISVTX);
3221 error = security_inode_mkdir(dir, dentry, mode);
3222 if (error)
3223 return error;
3225 if (max_links && dir->i_nlink >= max_links)
3226 return -EMLINK;
3228 error = dir->i_op->mkdir(dir, dentry, mode);
3229 if (!error)
3230 fsnotify_mkdir(dir, dentry);
3231 return error;
3234 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3236 struct dentry *dentry;
3237 struct path path;
3238 int error;
3239 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3241 retry:
3242 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3243 if (IS_ERR(dentry))
3244 return PTR_ERR(dentry);
3246 if (!IS_POSIXACL(path.dentry->d_inode))
3247 mode &= ~current_umask();
3248 error = security_path_mkdir(&path, dentry, mode);
3249 if (!error)
3250 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3251 done_path_create(&path, dentry);
3252 if (retry_estale(error, lookup_flags)) {
3253 lookup_flags |= LOOKUP_REVAL;
3254 goto retry;
3256 return error;
3259 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3261 return sys_mkdirat(AT_FDCWD, pathname, mode);
3265 * The dentry_unhash() helper will try to drop the dentry early: we
3266 * should have a usage count of 1 if we're the only user of this
3267 * dentry, and if that is true (possibly after pruning the dcache),
3268 * then we drop the dentry now.
3270 * A low-level filesystem can, if it choses, legally
3271 * do a
3273 * if (!d_unhashed(dentry))
3274 * return -EBUSY;
3276 * if it cannot handle the case of removing a directory
3277 * that is still in use by something else..
3279 void dentry_unhash(struct dentry *dentry)
3281 shrink_dcache_parent(dentry);
3282 spin_lock(&dentry->d_lock);
3283 if (dentry->d_count == 1)
3284 __d_drop(dentry);
3285 spin_unlock(&dentry->d_lock);
3288 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3290 int error = may_delete(dir, dentry, 1);
3292 if (error)
3293 return error;
3295 if (!dir->i_op->rmdir)
3296 return -EPERM;
3298 dget(dentry);
3299 mutex_lock(&dentry->d_inode->i_mutex);
3301 error = -EBUSY;
3302 if (d_mountpoint(dentry))
3303 goto out;
3305 error = security_inode_rmdir(dir, dentry);
3306 if (error)
3307 goto out;
3309 shrink_dcache_parent(dentry);
3310 error = dir->i_op->rmdir(dir, dentry);
3311 if (error)
3312 goto out;
3314 dentry->d_inode->i_flags |= S_DEAD;
3315 dont_mount(dentry);
3317 out:
3318 mutex_unlock(&dentry->d_inode->i_mutex);
3319 dput(dentry);
3320 if (!error)
3321 d_delete(dentry);
3322 return error;
3325 static long do_rmdir(int dfd, const char __user *pathname)
3327 int error = 0;
3328 struct filename *name;
3329 struct dentry *dentry;
3330 struct nameidata nd;
3331 unsigned int lookup_flags = 0;
3332 retry:
3333 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3334 if (IS_ERR(name))
3335 return PTR_ERR(name);
3337 switch(nd.last_type) {
3338 case LAST_DOTDOT:
3339 error = -ENOTEMPTY;
3340 goto exit1;
3341 case LAST_DOT:
3342 error = -EINVAL;
3343 goto exit1;
3344 case LAST_ROOT:
3345 error = -EBUSY;
3346 goto exit1;
3349 nd.flags &= ~LOOKUP_PARENT;
3350 error = mnt_want_write(nd.path.mnt);
3351 if (error)
3352 goto exit1;
3354 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3355 dentry = lookup_hash(&nd);
3356 error = PTR_ERR(dentry);
3357 if (IS_ERR(dentry))
3358 goto exit2;
3359 if (!dentry->d_inode) {
3360 error = -ENOENT;
3361 goto exit3;
3363 error = security_path_rmdir(&nd.path, dentry);
3364 if (error)
3365 goto exit3;
3366 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3367 exit3:
3368 dput(dentry);
3369 exit2:
3370 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3371 mnt_drop_write(nd.path.mnt);
3372 exit1:
3373 path_put(&nd.path);
3374 putname(name);
3375 if (retry_estale(error, lookup_flags)) {
3376 lookup_flags |= LOOKUP_REVAL;
3377 goto retry;
3379 return error;
3382 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3384 return do_rmdir(AT_FDCWD, pathname);
3387 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3389 int error = may_delete(dir, dentry, 0);
3391 if (error)
3392 return error;
3394 if (!dir->i_op->unlink)
3395 return -EPERM;
3397 mutex_lock(&dentry->d_inode->i_mutex);
3398 if (d_mountpoint(dentry))
3399 error = -EBUSY;
3400 else {
3401 error = security_inode_unlink(dir, dentry);
3402 if (!error) {
3403 error = dir->i_op->unlink(dir, dentry);
3404 if (!error)
3405 dont_mount(dentry);
3408 mutex_unlock(&dentry->d_inode->i_mutex);
3410 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3411 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3412 fsnotify_link_count(dentry->d_inode);
3413 d_delete(dentry);
3416 return error;
3420 * Make sure that the actual truncation of the file will occur outside its
3421 * directory's i_mutex. Truncate can take a long time if there is a lot of
3422 * writeout happening, and we don't want to prevent access to the directory
3423 * while waiting on the I/O.
3425 static long do_unlinkat(int dfd, const char __user *pathname)
3427 int error;
3428 struct filename *name;
3429 struct dentry *dentry;
3430 struct nameidata nd;
3431 struct inode *inode = NULL;
3432 unsigned int lookup_flags = 0;
3433 retry:
3434 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3435 if (IS_ERR(name))
3436 return PTR_ERR(name);
3438 error = -EISDIR;
3439 if (nd.last_type != LAST_NORM)
3440 goto exit1;
3442 nd.flags &= ~LOOKUP_PARENT;
3443 error = mnt_want_write(nd.path.mnt);
3444 if (error)
3445 goto exit1;
3447 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3448 dentry = lookup_hash(&nd);
3449 error = PTR_ERR(dentry);
3450 if (!IS_ERR(dentry)) {
3451 /* Why not before? Because we want correct error value */
3452 if (nd.last.name[nd.last.len])
3453 goto slashes;
3454 inode = dentry->d_inode;
3455 if (!inode)
3456 goto slashes;
3457 ihold(inode);
3458 error = security_path_unlink(&nd.path, dentry);
3459 if (error)
3460 goto exit2;
3461 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3462 exit2:
3463 dput(dentry);
3465 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3466 if (inode)
3467 iput(inode); /* truncate the inode here */
3468 mnt_drop_write(nd.path.mnt);
3469 exit1:
3470 path_put(&nd.path);
3471 putname(name);
3472 if (retry_estale(error, lookup_flags)) {
3473 lookup_flags |= LOOKUP_REVAL;
3474 inode = NULL;
3475 goto retry;
3477 return error;
3479 slashes:
3480 error = !dentry->d_inode ? -ENOENT :
3481 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3482 goto exit2;
3485 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3487 if ((flag & ~AT_REMOVEDIR) != 0)
3488 return -EINVAL;
3490 if (flag & AT_REMOVEDIR)
3491 return do_rmdir(dfd, pathname);
3493 return do_unlinkat(dfd, pathname);
3496 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3498 return do_unlinkat(AT_FDCWD, pathname);
3501 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3503 int error = may_create(dir, dentry);
3505 if (error)
3506 return error;
3508 if (!dir->i_op->symlink)
3509 return -EPERM;
3511 error = security_inode_symlink(dir, dentry, oldname);
3512 if (error)
3513 return error;
3515 error = dir->i_op->symlink(dir, dentry, oldname);
3516 if (!error)
3517 fsnotify_create(dir, dentry);
3518 return error;
3521 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3522 int, newdfd, const char __user *, newname)
3524 int error;
3525 struct filename *from;
3526 struct dentry *dentry;
3527 struct path path;
3528 unsigned int lookup_flags = 0;
3530 from = getname(oldname);
3531 if (IS_ERR(from))
3532 return PTR_ERR(from);
3533 retry:
3534 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3535 error = PTR_ERR(dentry);
3536 if (IS_ERR(dentry))
3537 goto out_putname;
3539 error = security_path_symlink(&path, dentry, from->name);
3540 if (!error)
3541 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3542 done_path_create(&path, dentry);
3543 if (retry_estale(error, lookup_flags)) {
3544 lookup_flags |= LOOKUP_REVAL;
3545 goto retry;
3547 out_putname:
3548 putname(from);
3549 return error;
3552 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3554 return sys_symlinkat(oldname, AT_FDCWD, newname);
3557 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3559 struct inode *inode = old_dentry->d_inode;
3560 unsigned max_links = dir->i_sb->s_max_links;
3561 int error;
3563 if (!inode)
3564 return -ENOENT;
3566 error = may_create(dir, new_dentry);
3567 if (error)
3568 return error;
3570 if (dir->i_sb != inode->i_sb)
3571 return -EXDEV;
3574 * A link to an append-only or immutable file cannot be created.
3576 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3577 return -EPERM;
3578 if (!dir->i_op->link)
3579 return -EPERM;
3580 if (S_ISDIR(inode->i_mode))
3581 return -EPERM;
3583 error = security_inode_link(old_dentry, dir, new_dentry);
3584 if (error)
3585 return error;
3587 mutex_lock(&inode->i_mutex);
3588 /* Make sure we don't allow creating hardlink to an unlinked file */
3589 if (inode->i_nlink == 0)
3590 error = -ENOENT;
3591 else if (max_links && inode->i_nlink >= max_links)
3592 error = -EMLINK;
3593 else
3594 error = dir->i_op->link(old_dentry, dir, new_dentry);
3595 mutex_unlock(&inode->i_mutex);
3596 if (!error)
3597 fsnotify_link(dir, inode, new_dentry);
3598 return error;
3602 * Hardlinks are often used in delicate situations. We avoid
3603 * security-related surprises by not following symlinks on the
3604 * newname. --KAB
3606 * We don't follow them on the oldname either to be compatible
3607 * with linux 2.0, and to avoid hard-linking to directories
3608 * and other special files. --ADM
3610 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3611 int, newdfd, const char __user *, newname, int, flags)
3613 struct dentry *new_dentry;
3614 struct path old_path, new_path;
3615 int how = 0;
3616 int error;
3618 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3619 return -EINVAL;
3621 * To use null names we require CAP_DAC_READ_SEARCH
3622 * This ensures that not everyone will be able to create
3623 * handlink using the passed filedescriptor.
3625 if (flags & AT_EMPTY_PATH) {
3626 if (!capable(CAP_DAC_READ_SEARCH))
3627 return -ENOENT;
3628 how = LOOKUP_EMPTY;
3631 if (flags & AT_SYMLINK_FOLLOW)
3632 how |= LOOKUP_FOLLOW;
3633 retry:
3634 error = user_path_at(olddfd, oldname, how, &old_path);
3635 if (error)
3636 return error;
3638 new_dentry = user_path_create(newdfd, newname, &new_path,
3639 (how & LOOKUP_REVAL));
3640 error = PTR_ERR(new_dentry);
3641 if (IS_ERR(new_dentry))
3642 goto out;
3644 error = -EXDEV;
3645 if (old_path.mnt != new_path.mnt)
3646 goto out_dput;
3647 error = may_linkat(&old_path);
3648 if (unlikely(error))
3649 goto out_dput;
3650 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3651 if (error)
3652 goto out_dput;
3653 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3654 out_dput:
3655 done_path_create(&new_path, new_dentry);
3656 if (retry_estale(error, how)) {
3657 how |= LOOKUP_REVAL;
3658 goto retry;
3660 out:
3661 path_put(&old_path);
3663 return error;
3666 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3668 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3672 * The worst of all namespace operations - renaming directory. "Perverted"
3673 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3674 * Problems:
3675 * a) we can get into loop creation. Check is done in is_subdir().
3676 * b) race potential - two innocent renames can create a loop together.
3677 * That's where 4.4 screws up. Current fix: serialization on
3678 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3679 * story.
3680 * c) we have to lock _three_ objects - parents and victim (if it exists).
3681 * And that - after we got ->i_mutex on parents (until then we don't know
3682 * whether the target exists). Solution: try to be smart with locking
3683 * order for inodes. We rely on the fact that tree topology may change
3684 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3685 * move will be locked. Thus we can rank directories by the tree
3686 * (ancestors first) and rank all non-directories after them.
3687 * That works since everybody except rename does "lock parent, lookup,
3688 * lock child" and rename is under ->s_vfs_rename_mutex.
3689 * HOWEVER, it relies on the assumption that any object with ->lookup()
3690 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3691 * we'd better make sure that there's no link(2) for them.
3692 * d) conversion from fhandle to dentry may come in the wrong moment - when
3693 * we are removing the target. Solution: we will have to grab ->i_mutex
3694 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3695 * ->i_mutex on parents, which works but leads to some truly excessive
3696 * locking].
3698 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3699 struct inode *new_dir, struct dentry *new_dentry)
3701 int error = 0;
3702 struct inode *target = new_dentry->d_inode;
3703 unsigned max_links = new_dir->i_sb->s_max_links;
3706 * If we are going to change the parent - check write permissions,
3707 * we'll need to flip '..'.
3709 if (new_dir != old_dir) {
3710 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3711 if (error)
3712 return error;
3715 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3716 if (error)
3717 return error;
3719 dget(new_dentry);
3720 if (target)
3721 mutex_lock(&target->i_mutex);
3723 error = -EBUSY;
3724 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3725 goto out;
3727 error = -EMLINK;
3728 if (max_links && !target && new_dir != old_dir &&
3729 new_dir->i_nlink >= max_links)
3730 goto out;
3732 if (target)
3733 shrink_dcache_parent(new_dentry);
3734 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3735 if (error)
3736 goto out;
3738 if (target) {
3739 target->i_flags |= S_DEAD;
3740 dont_mount(new_dentry);
3742 out:
3743 if (target)
3744 mutex_unlock(&target->i_mutex);
3745 dput(new_dentry);
3746 if (!error)
3747 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3748 d_move(old_dentry,new_dentry);
3749 return error;
3752 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3753 struct inode *new_dir, struct dentry *new_dentry)
3755 struct inode *target = new_dentry->d_inode;
3756 int error;
3758 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3759 if (error)
3760 return error;
3762 dget(new_dentry);
3763 if (target)
3764 mutex_lock(&target->i_mutex);
3766 error = -EBUSY;
3767 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3768 goto out;
3770 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3771 if (error)
3772 goto out;
3774 if (target)
3775 dont_mount(new_dentry);
3776 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3777 d_move(old_dentry, new_dentry);
3778 out:
3779 if (target)
3780 mutex_unlock(&target->i_mutex);
3781 dput(new_dentry);
3782 return error;
3785 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3786 struct inode *new_dir, struct dentry *new_dentry)
3788 int error;
3789 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3790 const unsigned char *old_name;
3792 if (old_dentry->d_inode == new_dentry->d_inode)
3793 return 0;
3795 error = may_delete(old_dir, old_dentry, is_dir);
3796 if (error)
3797 return error;
3799 if (!new_dentry->d_inode)
3800 error = may_create(new_dir, new_dentry);
3801 else
3802 error = may_delete(new_dir, new_dentry, is_dir);
3803 if (error)
3804 return error;
3806 if (!old_dir->i_op->rename)
3807 return -EPERM;
3809 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3811 if (is_dir)
3812 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3813 else
3814 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3815 if (!error)
3816 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3817 new_dentry->d_inode, old_dentry);
3818 fsnotify_oldname_free(old_name);
3820 return error;
3823 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3824 int, newdfd, const char __user *, newname)
3826 struct dentry *old_dir, *new_dir;
3827 struct dentry *old_dentry, *new_dentry;
3828 struct dentry *trap;
3829 struct nameidata oldnd, newnd;
3830 struct filename *from;
3831 struct filename *to;
3832 unsigned int lookup_flags = 0;
3833 bool should_retry = false;
3834 int error;
3835 retry:
3836 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
3837 if (IS_ERR(from)) {
3838 error = PTR_ERR(from);
3839 goto exit;
3842 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
3843 if (IS_ERR(to)) {
3844 error = PTR_ERR(to);
3845 goto exit1;
3848 error = -EXDEV;
3849 if (oldnd.path.mnt != newnd.path.mnt)
3850 goto exit2;
3852 old_dir = oldnd.path.dentry;
3853 error = -EBUSY;
3854 if (oldnd.last_type != LAST_NORM)
3855 goto exit2;
3857 new_dir = newnd.path.dentry;
3858 if (newnd.last_type != LAST_NORM)
3859 goto exit2;
3861 error = mnt_want_write(oldnd.path.mnt);
3862 if (error)
3863 goto exit2;
3865 oldnd.flags &= ~LOOKUP_PARENT;
3866 newnd.flags &= ~LOOKUP_PARENT;
3867 newnd.flags |= LOOKUP_RENAME_TARGET;
3869 trap = lock_rename(new_dir, old_dir);
3871 old_dentry = lookup_hash(&oldnd);
3872 error = PTR_ERR(old_dentry);
3873 if (IS_ERR(old_dentry))
3874 goto exit3;
3875 /* source must exist */
3876 error = -ENOENT;
3877 if (!old_dentry->d_inode)
3878 goto exit4;
3879 /* unless the source is a directory trailing slashes give -ENOTDIR */
3880 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3881 error = -ENOTDIR;
3882 if (oldnd.last.name[oldnd.last.len])
3883 goto exit4;
3884 if (newnd.last.name[newnd.last.len])
3885 goto exit4;
3887 /* source should not be ancestor of target */
3888 error = -EINVAL;
3889 if (old_dentry == trap)
3890 goto exit4;
3891 new_dentry = lookup_hash(&newnd);
3892 error = PTR_ERR(new_dentry);
3893 if (IS_ERR(new_dentry))
3894 goto exit4;
3895 /* target should not be an ancestor of source */
3896 error = -ENOTEMPTY;
3897 if (new_dentry == trap)
3898 goto exit5;
3900 error = security_path_rename(&oldnd.path, old_dentry,
3901 &newnd.path, new_dentry);
3902 if (error)
3903 goto exit5;
3904 error = vfs_rename(old_dir->d_inode, old_dentry,
3905 new_dir->d_inode, new_dentry);
3906 exit5:
3907 dput(new_dentry);
3908 exit4:
3909 dput(old_dentry);
3910 exit3:
3911 unlock_rename(new_dir, old_dir);
3912 mnt_drop_write(oldnd.path.mnt);
3913 exit2:
3914 if (retry_estale(error, lookup_flags))
3915 should_retry = true;
3916 path_put(&newnd.path);
3917 putname(to);
3918 exit1:
3919 path_put(&oldnd.path);
3920 putname(from);
3921 if (should_retry) {
3922 should_retry = false;
3923 lookup_flags |= LOOKUP_REVAL;
3924 goto retry;
3926 exit:
3927 return error;
3930 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3932 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3935 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3937 int len;
3939 len = PTR_ERR(link);
3940 if (IS_ERR(link))
3941 goto out;
3943 len = strlen(link);
3944 if (len > (unsigned) buflen)
3945 len = buflen;
3946 if (copy_to_user(buffer, link, len))
3947 len = -EFAULT;
3948 out:
3949 return len;
3953 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3954 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3955 * using) it for any given inode is up to filesystem.
3957 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3959 struct nameidata nd;
3960 void *cookie;
3961 int res;
3963 nd.depth = 0;
3964 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3965 if (IS_ERR(cookie))
3966 return PTR_ERR(cookie);
3968 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3969 if (dentry->d_inode->i_op->put_link)
3970 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3971 return res;
3974 int vfs_follow_link(struct nameidata *nd, const char *link)
3976 return __vfs_follow_link(nd, link);
3979 /* get the link contents into pagecache */
3980 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3982 char *kaddr;
3983 struct page *page;
3984 struct address_space *mapping = dentry->d_inode->i_mapping;
3985 page = read_mapping_page(mapping, 0, NULL);
3986 if (IS_ERR(page))
3987 return (char*)page;
3988 *ppage = page;
3989 kaddr = kmap(page);
3990 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3991 return kaddr;
3994 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3996 struct page *page = NULL;
3997 char *s = page_getlink(dentry, &page);
3998 int res = vfs_readlink(dentry,buffer,buflen,s);
3999 if (page) {
4000 kunmap(page);
4001 page_cache_release(page);
4003 return res;
4006 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4008 struct page *page = NULL;
4009 nd_set_link(nd, page_getlink(dentry, &page));
4010 return page;
4013 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4015 struct page *page = cookie;
4017 if (page) {
4018 kunmap(page);
4019 page_cache_release(page);
4024 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4026 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4028 struct address_space *mapping = inode->i_mapping;
4029 struct page *page;
4030 void *fsdata;
4031 int err;
4032 char *kaddr;
4033 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4034 if (nofs)
4035 flags |= AOP_FLAG_NOFS;
4037 retry:
4038 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4039 flags, &page, &fsdata);
4040 if (err)
4041 goto fail;
4043 kaddr = kmap_atomic(page);
4044 memcpy(kaddr, symname, len-1);
4045 kunmap_atomic(kaddr);
4047 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4048 page, fsdata);
4049 if (err < 0)
4050 goto fail;
4051 if (err < len-1)
4052 goto retry;
4054 mark_inode_dirty(inode);
4055 return 0;
4056 fail:
4057 return err;
4060 int page_symlink(struct inode *inode, const char *symname, int len)
4062 return __page_symlink(inode, symname, len,
4063 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4066 const struct inode_operations page_symlink_inode_operations = {
4067 .readlink = generic_readlink,
4068 .follow_link = page_follow_link_light,
4069 .put_link = page_put_link,
4072 EXPORT_SYMBOL(user_path_at);
4073 EXPORT_SYMBOL(follow_down_one);
4074 EXPORT_SYMBOL(follow_down);
4075 EXPORT_SYMBOL(follow_up);
4076 EXPORT_SYMBOL(get_write_access); /* nfsd */
4077 EXPORT_SYMBOL(lock_rename);
4078 EXPORT_SYMBOL(lookup_one_len);
4079 EXPORT_SYMBOL(page_follow_link_light);
4080 EXPORT_SYMBOL(page_put_link);
4081 EXPORT_SYMBOL(page_readlink);
4082 EXPORT_SYMBOL(__page_symlink);
4083 EXPORT_SYMBOL(page_symlink);
4084 EXPORT_SYMBOL(page_symlink_inode_operations);
4085 EXPORT_SYMBOL(kern_path);
4086 EXPORT_SYMBOL(vfs_path_lookup);
4087 EXPORT_SYMBOL(inode_permission);
4088 EXPORT_SYMBOL(unlock_rename);
4089 EXPORT_SYMBOL(vfs_create);
4090 EXPORT_SYMBOL(vfs_follow_link);
4091 EXPORT_SYMBOL(vfs_link);
4092 EXPORT_SYMBOL(vfs_mkdir);
4093 EXPORT_SYMBOL(vfs_mknod);
4094 EXPORT_SYMBOL(generic_permission);
4095 EXPORT_SYMBOL(vfs_readlink);
4096 EXPORT_SYMBOL(vfs_rename);
4097 EXPORT_SYMBOL(vfs_rmdir);
4098 EXPORT_SYMBOL(vfs_symlink);
4099 EXPORT_SYMBOL(vfs_unlink);
4100 EXPORT_SYMBOL(dentry_unhash);
4101 EXPORT_SYMBOL(generic_readlink);