blob: 5008f01787f5681730a9273693a1a863e7e6cc7e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070035#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37
Eric Parise81e3f42009-12-04 15:47:36 -050038#include "internal.h"
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030074 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100140static char *getname_flags(const char __user *filename, int flags, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100151 if (retval == -ENOENT && empty)
152 *empty = 1;
Al Virof52e0c12011-03-14 18:56:51 -0400153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(tmp);
155 result = ERR_PTR(retval);
156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
158 }
159 audit_getname(result);
160 return result;
161}
162
Al Virof52e0c12011-03-14 18:56:51 -0400163char *getname(const char __user * filename)
164{
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100165 return getname_flags(filename, 0, 0);
Al Virof52e0c12011-03-14 18:56:51 -0400166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168#ifdef CONFIG_AUDITSYSCALL
169void putname(const char *name)
170{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400171 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 audit_putname(name);
173 else
174 __putname(name);
175}
176EXPORT_SYMBOL(putname);
177#endif
178
Linus Torvaldse77819e2011-07-22 19:30:19 -0700179static int check_acl(struct inode *inode, int mask)
180{
Linus Torvalds84635d62011-07-25 22:47:03 -0700181#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700182 struct posix_acl *acl;
183
Linus Torvaldse77819e2011-07-22 19:30:19 -0700184 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400185 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
186 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700187 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400188 /* no ->get_acl() calls in RCU mode... */
189 if (acl == ACL_NOT_CACHED)
190 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300191 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700192 }
193
194 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
195
196 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200197 * A filesystem can force a ACL callback by just never filling the
198 * ACL cache. But normally you'd fill the cache either at inode
199 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700200 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200201 * If the filesystem doesn't have a get_acl() function at all, we'll
202 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700203 */
204 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200205 if (inode->i_op->get_acl) {
206 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
207 if (IS_ERR(acl))
208 return PTR_ERR(acl);
209 } else {
210 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
211 return -EAGAIN;
212 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700213 }
214
215 if (acl) {
216 int error = posix_acl_permission(inode, acl, mask);
217 posix_acl_release(acl);
218 return error;
219 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700220#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700221
222 return -EAGAIN;
223}
224
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700225/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530226 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700227 */
Al Viro7e401452011-06-20 19:12:17 -0400228static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700230 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700231
Serge E. Hallyne795b712011-03-23 16:43:25 -0700232 if (current_user_ns() != inode_userns(inode))
233 goto other_perms;
234
Linus Torvalds14067ff2011-07-25 19:55:52 -0700235 if (likely(current_fsuid() == inode->i_uid))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700236 mode >>= 6;
237 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400239 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100240 if (error != -EAGAIN)
241 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700242 }
243
244 if (in_group_p(inode->i_gid))
245 mode >>= 3;
246 }
247
Serge E. Hallyne795b712011-03-23 16:43:25 -0700248other_perms:
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700249 /*
250 * If the DACs are ok we don't need any capability check.
251 */
Al Viro9c2c7032011-06-20 19:06:22 -0400252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700253 return 0;
254 return -EACCES;
255}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100258 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100265 * are used for other things.
266 *
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 */
Al Viro2830ba72011-06-20 19:16:29 -0400271int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700273 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530276 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 */
Al Viro7e401452011-06-20 19:12:17 -0400278 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700279 if (ret != -EACCES)
280 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Al Virod594e7e2011-06-20 19:55:42 -0400282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
285 return 0;
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
288 return 0;
289 return -EACCES;
290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 /*
292 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 */
Al Virod594e7e2011-06-20 19:55:42 -0400296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 return 0;
299
300 /*
301 * Searching includes executable on directories, else just read.
302 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400304 if (mask == MAY_READ)
Serge E. Hallyne795b712011-03-23 16:43:25 -0700305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return 0;
307
308 return -EACCES;
309}
310
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700311/*
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
316 */
317static inline int do_inode_permission(struct inode *inode, int mask)
318{
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
322
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
327 }
328 return generic_permission(inode, mask);
329}
330
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200331/**
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200335 *
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530340 *
341 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200342 */
Al Virof419a2e2008-07-22 00:07:17 -0400343int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
Al Viroe6305c42008-07-15 21:03:57 -0400345 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700347 if (unlikely(mask & MAY_WRITE)) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700348 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 /*
351 * Nobody gets write access to a read-only fs.
352 */
353 if (IS_RDONLY(inode) &&
354 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
355 return -EROFS;
356
357 /*
358 * Nobody gets write access to an immutable file.
359 */
360 if (IS_IMMUTABLE(inode))
361 return -EACCES;
362 }
363
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700364 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (retval)
366 return retval;
367
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700368 retval = devcgroup_inode_permission(inode, mask);
369 if (retval)
370 return retval;
371
Eric Parisd09ca732010-07-23 11:43:57 -0400372 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
Al Virof4d6ff82011-06-19 13:14:21 -0400375/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800376 * path_get - get a reference to a path
377 * @path: path to get the reference to
378 *
379 * Given a path increment the reference count to the dentry and the vfsmount.
380 */
381void path_get(struct path *path)
382{
383 mntget(path->mnt);
384 dget(path->dentry);
385}
386EXPORT_SYMBOL(path_get);
387
388/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800389 * path_put - put a reference to a path
390 * @path: path to put the reference to
391 *
392 * Given a path decrement the reference count to the dentry and the vfsmount.
393 */
394void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
Jan Blunck1d957f92008-02-14 19:34:35 -0800396 dput(path->dentry);
397 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
Jan Blunck1d957f92008-02-14 19:34:35 -0800399EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Al Viro19660af2011-03-25 10:32:48 -0400401/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100402 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400403 * Documentation/filesystems/path-lookup.txt). In situations when we can't
404 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
405 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
406 * mode. Refcounts are grabbed at the last known good point before rcu-walk
407 * got stuck, so ref-walk may continue from there. If this is not successful
408 * (eg. a seqcount has changed), then failure is returned and it's up to caller
409 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100410 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100411
412/**
Al Viro19660af2011-03-25 10:32:48 -0400413 * unlazy_walk - try to switch to ref-walk mode.
414 * @nd: nameidata pathwalk data
415 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800416 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100417 *
Al Viro19660af2011-03-25 10:32:48 -0400418 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
419 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
420 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100421 */
Al Viro19660af2011-03-25 10:32:48 -0400422static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100423{
424 struct fs_struct *fs = current->fs;
425 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500426 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100427
428 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500429 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
430 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100431 spin_lock(&fs->lock);
432 if (nd->root.mnt != fs->root.mnt ||
433 nd->root.dentry != fs->root.dentry)
434 goto err_root;
435 }
436 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400437 if (!dentry) {
438 if (!__d_rcu_to_refcount(parent, nd->seq))
439 goto err_parent;
440 BUG_ON(nd->inode != parent->d_inode);
441 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400442 if (dentry->d_parent != parent)
443 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400444 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
445 if (!__d_rcu_to_refcount(dentry, nd->seq))
446 goto err_child;
447 /*
448 * If the sequence check on the child dentry passed, then
449 * the child has not been removed from its parent. This
450 * means the parent dentry must be valid and able to take
451 * a reference at this point.
452 */
453 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
454 BUG_ON(!parent->d_count);
455 parent->d_count++;
456 spin_unlock(&dentry->d_lock);
457 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100458 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500459 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100460 path_get(&nd->root);
461 spin_unlock(&fs->lock);
462 }
463 mntget(nd->path.mnt);
464
465 rcu_read_unlock();
466 br_read_unlock(vfsmount_lock);
467 nd->flags &= ~LOOKUP_RCU;
468 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400469
470err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100471 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400472err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100473 spin_unlock(&parent->d_lock);
474err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500475 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100476 spin_unlock(&fs->lock);
477 return -ECHILD;
478}
479
Nick Piggin31e6b012011-01-07 17:49:52 +1100480/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700481 * release_open_intent - free up open intent resources
482 * @nd: pointer to nameidata
483 */
484void release_open_intent(struct nameidata *nd)
485{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800486 struct file *file = nd->intent.open.file;
487
488 if (file && !IS_ERR(file)) {
489 if (file->f_path.dentry == NULL)
490 put_filp(file);
491 else
492 fput(file);
493 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700494}
495
Al Virof60aef72011-02-15 01:35:28 -0500496static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100497{
Al Virof60aef72011-02-15 01:35:28 -0500498 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100499}
500
Al Viro9f1fafe2011-03-25 11:00:12 -0400501/**
502 * complete_walk - successful completion of path walk
503 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500504 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400505 * If we had been in RCU mode, drop out of it and legitimize nd->path.
506 * Revalidate the final result, unless we'd already done that during
507 * the path walk or the filesystem doesn't ask for it. Return 0 on
508 * success, -error on failure. In case of failure caller does not
509 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500510 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400511static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500512{
Al Viro16c2cd72011-02-22 15:50:10 -0500513 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500514 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500515
Al Viro9f1fafe2011-03-25 11:00:12 -0400516 if (nd->flags & LOOKUP_RCU) {
517 nd->flags &= ~LOOKUP_RCU;
518 if (!(nd->flags & LOOKUP_ROOT))
519 nd->root.mnt = NULL;
520 spin_lock(&dentry->d_lock);
521 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
522 spin_unlock(&dentry->d_lock);
523 rcu_read_unlock();
524 br_read_unlock(vfsmount_lock);
525 return -ECHILD;
526 }
527 BUG_ON(nd->inode != dentry->d_inode);
528 spin_unlock(&dentry->d_lock);
529 mntget(nd->path.mnt);
530 rcu_read_unlock();
531 br_read_unlock(vfsmount_lock);
532 }
533
Al Viro16c2cd72011-02-22 15:50:10 -0500534 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500535 return 0;
536
Al Viro16c2cd72011-02-22 15:50:10 -0500537 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
538 return 0;
539
540 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
541 return 0;
542
543 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100544 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500545 if (status > 0)
546 return 0;
547
Al Viro16c2cd72011-02-22 15:50:10 -0500548 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500549 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500550
Al Viro9f1fafe2011-03-25 11:00:12 -0400551 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500552 return status;
553}
554
Al Viro2a737872009-04-07 11:49:53 -0400555static __always_inline void set_root(struct nameidata *nd)
556{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200557 if (!nd->root.mnt)
558 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400559}
560
Al Viro6de88d72009-08-09 01:41:57 +0400561static int link_path_walk(const char *, struct nameidata *);
562
Nick Piggin31e6b012011-01-07 17:49:52 +1100563static __always_inline void set_root_rcu(struct nameidata *nd)
564{
565 if (!nd->root.mnt) {
566 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100567 unsigned seq;
568
569 do {
570 seq = read_seqcount_begin(&fs->seq);
571 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700572 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100573 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100574 }
575}
576
Arjan van de Venf1662352006-01-14 13:21:31 -0800577static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Nick Piggin31e6b012011-01-07 17:49:52 +1100579 int ret;
580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (IS_ERR(link))
582 goto fail;
583
584 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400585 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800586 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400587 nd->path = nd->root;
588 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500589 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100591 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100592
Nick Piggin31e6b012011-01-07 17:49:52 +1100593 ret = link_path_walk(link, nd);
594 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800596 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return PTR_ERR(link);
598}
599
Jan Blunck1d957f92008-02-14 19:34:35 -0800600static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700601{
602 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800603 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700604 mntput(path->mnt);
605}
606
Nick Piggin7b9337a2011-01-14 08:42:43 +0000607static inline void path_to_nameidata(const struct path *path,
608 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700609{
Nick Piggin31e6b012011-01-07 17:49:52 +1100610 if (!(nd->flags & LOOKUP_RCU)) {
611 dput(nd->path.dentry);
612 if (nd->path.mnt != path->mnt)
613 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800614 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100615 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800616 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700617}
618
Al Viro574197e2011-03-14 22:20:34 -0400619static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
620{
621 struct inode *inode = link->dentry->d_inode;
622 if (!IS_ERR(cookie) && inode->i_op->put_link)
623 inode->i_op->put_link(link->dentry, nd, cookie);
624 path_put(link);
625}
626
Al Virodef4af32009-12-26 08:37:05 -0500627static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400628follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800629{
630 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000631 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800632
Al Viro844a3912011-02-15 00:38:26 -0500633 BUG_ON(nd->flags & LOOKUP_RCU);
634
Al Viro0e794582011-03-16 02:45:02 -0400635 if (link->mnt == nd->path.mnt)
636 mntget(link->mnt);
637
Al Viro574197e2011-03-14 22:20:34 -0400638 if (unlikely(current->total_link_count >= 40)) {
639 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400640 path_put(&nd->path);
641 return -ELOOP;
642 }
643 cond_resched();
644 current->total_link_count++;
645
Nick Piggin7b9337a2011-01-14 08:42:43 +0000646 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800647 nd_set_link(nd, NULL);
648
Al Viro36f3b4f2011-02-22 21:24:38 -0500649 error = security_inode_follow_link(link->dentry, nd);
650 if (error) {
651 *p = ERR_PTR(error); /* no ->put_link(), please */
652 path_put(&nd->path);
653 return error;
654 }
655
Al Viro86acdca12009-12-22 23:45:11 -0500656 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500657 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
658 error = PTR_ERR(*p);
659 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800660 char *s = nd_get_link(nd);
661 error = 0;
662 if (s)
663 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400664 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500665 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400666 nd->inode = nd->path.dentry->d_inode;
667 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400668 /* stepped on a _really_ weird one */
669 path_put(&nd->path);
670 error = -ELOOP;
671 }
672 }
Ian Kent051d3812006-03-27 01:14:53 -0800673 }
Ian Kent051d3812006-03-27 01:14:53 -0800674 return error;
675}
676
Nick Piggin31e6b012011-01-07 17:49:52 +1100677static int follow_up_rcu(struct path *path)
678{
679 struct vfsmount *parent;
680 struct dentry *mountpoint;
681
682 parent = path->mnt->mnt_parent;
683 if (parent == path->mnt)
684 return 0;
685 mountpoint = path->mnt->mnt_mountpoint;
686 path->dentry = mountpoint;
687 path->mnt = parent;
688 return 1;
689}
690
Al Virobab77eb2009-04-18 03:26:48 -0400691int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
693 struct vfsmount *parent;
694 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000695
696 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400697 parent = path->mnt->mnt_parent;
698 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000699 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701 }
702 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400703 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000704 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400705 dput(path->dentry);
706 path->dentry = mountpoint;
707 mntput(path->mnt);
708 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 return 1;
710}
711
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100712/*
David Howells9875cf82011-01-14 18:45:21 +0000713 * Perform an automount
714 * - return -EISDIR to tell follow_managed() to stop and return the path we
715 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
David Howells9875cf82011-01-14 18:45:21 +0000717static int follow_automount(struct path *path, unsigned flags,
718 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100719{
David Howells9875cf82011-01-14 18:45:21 +0000720 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000721 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100722
David Howells9875cf82011-01-14 18:45:21 +0000723 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
724 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700725
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200726 /* We don't want to mount if someone's just doing a stat -
727 * unless they're stat'ing a directory and appended a '/' to
728 * the name.
729 *
730 * We do, however, want to mount if someone wants to open or
731 * create a file of any type under the mountpoint, wants to
732 * traverse through the mountpoint or wants to open the
733 * mounted directory. Also, autofs may mark negative dentries
734 * as being automount points. These will need the attentions
735 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000736 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200737 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700738 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200739 path->dentry->d_inode)
740 return -EISDIR;
741
David Howells9875cf82011-01-14 18:45:21 +0000742 current->total_link_count++;
743 if (current->total_link_count >= 40)
744 return -ELOOP;
745
746 mnt = path->dentry->d_op->d_automount(path);
747 if (IS_ERR(mnt)) {
748 /*
749 * The filesystem is allowed to return -EISDIR here to indicate
750 * it doesn't want to automount. For instance, autofs would do
751 * this so that its userspace daemon can mount on this dentry.
752 *
753 * However, we can only permit this if it's a terminal point in
754 * the path being looked up; if it wasn't then the remainder of
755 * the path is inaccessible and we should say so.
756 */
Al Viro49084c32011-06-25 21:59:52 -0400757 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000758 return -EREMOTE;
759 return PTR_ERR(mnt);
760 }
David Howellsea5b7782011-01-14 19:10:03 +0000761
David Howells9875cf82011-01-14 18:45:21 +0000762 if (!mnt) /* mount collision */
763 return 0;
764
Al Viro8aef1882011-06-16 15:10:06 +0100765 if (!*need_mntput) {
766 /* lock_mount() may release path->mnt on error */
767 mntget(path->mnt);
768 *need_mntput = true;
769 }
Al Viro19a167a2011-01-17 01:35:23 -0500770 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000771
David Howellsea5b7782011-01-14 19:10:03 +0000772 switch (err) {
773 case -EBUSY:
774 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500775 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000776 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100777 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000778 path->mnt = mnt;
779 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000780 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500781 default:
782 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000783 }
Al Viro19a167a2011-01-17 01:35:23 -0500784
David Howells9875cf82011-01-14 18:45:21 +0000785}
786
787/*
788 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000789 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000790 * - Flagged as mountpoint
791 * - Flagged as automount point
792 *
793 * This may only be called in refwalk mode.
794 *
795 * Serialization is taken care of in namespace.c
796 */
797static int follow_managed(struct path *path, unsigned flags)
798{
Al Viro8aef1882011-06-16 15:10:06 +0100799 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000800 unsigned managed;
801 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100802 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000803
804 /* Given that we're not holding a lock here, we retain the value in a
805 * local variable for each dentry as we look at it so that we don't see
806 * the components of that value change under us */
807 while (managed = ACCESS_ONCE(path->dentry->d_flags),
808 managed &= DCACHE_MANAGED_DENTRY,
809 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000810 /* Allow the filesystem to manage the transit without i_mutex
811 * being held. */
812 if (managed & DCACHE_MANAGE_TRANSIT) {
813 BUG_ON(!path->dentry->d_op);
814 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400815 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000816 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100817 break;
David Howellscc53ce52011-01-14 18:45:26 +0000818 }
819
David Howells9875cf82011-01-14 18:45:21 +0000820 /* Transit to a mounted filesystem. */
821 if (managed & DCACHE_MOUNTED) {
822 struct vfsmount *mounted = lookup_mnt(path);
823 if (mounted) {
824 dput(path->dentry);
825 if (need_mntput)
826 mntput(path->mnt);
827 path->mnt = mounted;
828 path->dentry = dget(mounted->mnt_root);
829 need_mntput = true;
830 continue;
831 }
832
833 /* Something is mounted on this dentry in another
834 * namespace and/or whatever was mounted there in this
835 * namespace got unmounted before we managed to get the
836 * vfsmount_lock */
837 }
838
839 /* Handle an automount point */
840 if (managed & DCACHE_NEED_AUTOMOUNT) {
841 ret = follow_automount(path, flags, &need_mntput);
842 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100843 break;
David Howells9875cf82011-01-14 18:45:21 +0000844 continue;
845 }
846
847 /* We didn't change the current path point */
848 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
Al Viro8aef1882011-06-16 15:10:06 +0100850
851 if (need_mntput && path->mnt == mnt)
852 mntput(path->mnt);
853 if (ret == -EISDIR)
854 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +0000855 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
David Howellscc53ce52011-01-14 18:45:26 +0000858int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 struct vfsmount *mounted;
861
Al Viro1c755af2009-04-18 14:06:57 -0400862 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400864 dput(path->dentry);
865 mntput(path->mnt);
866 path->mnt = mounted;
867 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 return 1;
869 }
870 return 0;
871}
872
Ian Kent62a73752011-03-25 01:51:02 +0800873static inline bool managed_dentry_might_block(struct dentry *dentry)
874{
875 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
876 dentry->d_op->d_manage(dentry, true) < 0);
877}
878
David Howells9875cf82011-01-14 18:45:21 +0000879/*
Al Viro287548e2011-05-27 06:50:06 -0400880 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
881 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000882 */
883static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400884 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000885{
Ian Kent62a73752011-03-25 01:51:02 +0800886 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +0000887 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800888 /*
889 * Don't forget we might have a non-mountpoint managed dentry
890 * that wants to block transit.
891 */
Al Viro287548e2011-05-27 06:50:06 -0400892 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000893 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800894
895 if (!d_mountpoint(path->dentry))
896 break;
897
David Howells9875cf82011-01-14 18:45:21 +0000898 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
899 if (!mounted)
900 break;
901 path->mnt = mounted;
902 path->dentry = mounted->mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +0000903 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +0000904 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700905 /*
906 * Update the inode too. We don't need to re-check the
907 * dentry sequence number here after this d_inode read,
908 * because a mount-point is always pinned.
909 */
910 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000911 }
David Howells9875cf82011-01-14 18:45:21 +0000912 return true;
913}
914
Al Virodea39372011-05-27 06:53:39 -0400915static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400916{
Al Virodea39372011-05-27 06:53:39 -0400917 while (d_mountpoint(nd->path.dentry)) {
Al Viro287548e2011-05-27 06:50:06 -0400918 struct vfsmount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400919 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400920 if (!mounted)
921 break;
Al Virodea39372011-05-27 06:53:39 -0400922 nd->path.mnt = mounted;
923 nd->path.dentry = mounted->mnt_root;
924 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400925 }
926}
927
Nick Piggin31e6b012011-01-07 17:49:52 +1100928static int follow_dotdot_rcu(struct nameidata *nd)
929{
Nick Piggin31e6b012011-01-07 17:49:52 +1100930 set_root_rcu(nd);
931
David Howells9875cf82011-01-14 18:45:21 +0000932 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100933 if (nd->path.dentry == nd->root.dentry &&
934 nd->path.mnt == nd->root.mnt) {
935 break;
936 }
937 if (nd->path.dentry != nd->path.mnt->mnt_root) {
938 struct dentry *old = nd->path.dentry;
939 struct dentry *parent = old->d_parent;
940 unsigned seq;
941
942 seq = read_seqcount_begin(&parent->d_seq);
943 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500944 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100945 nd->path.dentry = parent;
946 nd->seq = seq;
947 break;
948 }
949 if (!follow_up_rcu(&nd->path))
950 break;
951 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100952 }
Al Virodea39372011-05-27 06:53:39 -0400953 follow_mount_rcu(nd);
954 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100955 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500956
957failed:
958 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500959 if (!(nd->flags & LOOKUP_ROOT))
960 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500961 rcu_read_unlock();
962 br_read_unlock(vfsmount_lock);
963 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100964}
965
David Howells9875cf82011-01-14 18:45:21 +0000966/*
David Howellscc53ce52011-01-14 18:45:26 +0000967 * Follow down to the covering mount currently visible to userspace. At each
968 * point, the filesystem owning that dentry may be queried as to whether the
969 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000970 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400971int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000972{
973 unsigned managed;
974 int ret;
975
976 while (managed = ACCESS_ONCE(path->dentry->d_flags),
977 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
978 /* Allow the filesystem to manage the transit without i_mutex
979 * being held.
980 *
981 * We indicate to the filesystem if someone is trying to mount
982 * something here. This gives autofs the chance to deny anyone
983 * other than its daemon the right to mount on its
984 * superstructure.
985 *
986 * The filesystem may sleep at this point.
987 */
988 if (managed & DCACHE_MANAGE_TRANSIT) {
989 BUG_ON(!path->dentry->d_op);
990 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000991 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400992 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000993 if (ret < 0)
994 return ret == -EISDIR ? 0 : ret;
995 }
996
997 /* Transit to a mounted filesystem. */
998 if (managed & DCACHE_MOUNTED) {
999 struct vfsmount *mounted = lookup_mnt(path);
1000 if (!mounted)
1001 break;
1002 dput(path->dentry);
1003 mntput(path->mnt);
1004 path->mnt = mounted;
1005 path->dentry = dget(mounted->mnt_root);
1006 continue;
1007 }
1008
1009 /* Don't handle automount points here */
1010 break;
1011 }
1012 return 0;
1013}
1014
1015/*
David Howells9875cf82011-01-14 18:45:21 +00001016 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1017 */
1018static void follow_mount(struct path *path)
1019{
1020 while (d_mountpoint(path->dentry)) {
1021 struct vfsmount *mounted = lookup_mnt(path);
1022 if (!mounted)
1023 break;
1024 dput(path->dentry);
1025 mntput(path->mnt);
1026 path->mnt = mounted;
1027 path->dentry = dget(mounted->mnt_root);
1028 }
1029}
1030
Nick Piggin31e6b012011-01-07 17:49:52 +11001031static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
Al Viro2a737872009-04-07 11:49:53 -04001033 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001036 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Al Viro2a737872009-04-07 11:49:53 -04001038 if (nd->path.dentry == nd->root.dentry &&
1039 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 break;
1041 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001042 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001043 /* rare case of legitimate dget_parent()... */
1044 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 dput(old);
1046 break;
1047 }
Al Viro3088dd72010-01-30 15:47:29 -05001048 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Al Viro79ed0222009-04-18 13:59:41 -04001051 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001052 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053}
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001056 * Allocate a dentry with name and parent, and perform a parent
1057 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1058 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1059 * have verified that no child exists while under i_mutex.
1060 */
1061static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1062 struct qstr *name, struct nameidata *nd)
1063{
1064 struct inode *inode = parent->d_inode;
1065 struct dentry *dentry;
1066 struct dentry *old;
1067
1068 /* Don't create child dentry for a dead directory. */
1069 if (unlikely(IS_DEADDIR(inode)))
1070 return ERR_PTR(-ENOENT);
1071
1072 dentry = d_alloc(parent, name);
1073 if (unlikely(!dentry))
1074 return ERR_PTR(-ENOMEM);
1075
1076 old = inode->i_op->lookup(inode, dentry, nd);
1077 if (unlikely(old)) {
1078 dput(dentry);
1079 dentry = old;
1080 }
1081 return dentry;
1082}
1083
1084/*
Josef Bacik44396f42011-05-31 11:58:49 -04001085 * We already have a dentry, but require a lookup to be performed on the parent
1086 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1087 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1088 * child exists while under i_mutex.
1089 */
1090static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1091 struct nameidata *nd)
1092{
1093 struct inode *inode = parent->d_inode;
1094 struct dentry *old;
1095
1096 /* Don't create child dentry for a dead directory. */
1097 if (unlikely(IS_DEADDIR(inode)))
1098 return ERR_PTR(-ENOENT);
1099
1100 old = inode->i_op->lookup(inode, dentry, nd);
1101 if (unlikely(old)) {
1102 dput(dentry);
1103 dentry = old;
1104 }
1105 return dentry;
1106}
1107
1108/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 * It's more convoluted than I'd like it to be, but... it's still fairly
1110 * small and for now I'd prefer to have fast path as straight as possible.
1111 * It _is_ time-critical.
1112 */
1113static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001114 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115{
Jan Blunck4ac91372008-02-14 19:34:32 -08001116 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001117 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001118 int need_reval = 1;
1119 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001120 int err;
1121
Al Viro3cac2602009-08-13 18:27:43 +04001122 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001123 * Rename seqlock is not required here because in the off chance
1124 * of a false negative due to a concurrent rename, we're going to
1125 * do the non-racy lookup, below.
1126 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001127 if (nd->flags & LOOKUP_RCU) {
1128 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001129 *inode = nd->inode;
1130 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001131 if (!dentry)
1132 goto unlazy;
1133
Nick Piggin31e6b012011-01-07 17:49:52 +11001134 /* Memory barrier in read_seqcount_begin of child is enough */
1135 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1136 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001137 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001138
Al Viro24643082011-02-15 01:26:22 -05001139 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001140 status = d_revalidate(dentry, nd);
1141 if (unlikely(status <= 0)) {
1142 if (status != -ECHILD)
1143 need_reval = 0;
1144 goto unlazy;
1145 }
Al Viro24643082011-02-15 01:26:22 -05001146 }
Josef Bacik44396f42011-05-31 11:58:49 -04001147 if (unlikely(d_need_lookup(dentry)))
1148 goto unlazy;
Nick Piggin31e6b012011-01-07 17:49:52 +11001149 path->mnt = mnt;
1150 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001151 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1152 goto unlazy;
1153 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1154 goto unlazy;
1155 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001156unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001157 if (unlazy_walk(nd, dentry))
1158 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001159 } else {
1160 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001161 }
Al Viro5a18fff2011-03-11 04:44:53 -05001162
Josef Bacik44396f42011-05-31 11:58:49 -04001163 if (dentry && unlikely(d_need_lookup(dentry))) {
1164 dput(dentry);
1165 dentry = NULL;
1166 }
Al Viro5a18fff2011-03-11 04:44:53 -05001167retry:
1168 if (unlikely(!dentry)) {
1169 struct inode *dir = parent->d_inode;
1170 BUG_ON(nd->inode != dir);
1171
1172 mutex_lock(&dir->i_mutex);
1173 dentry = d_lookup(parent, name);
1174 if (likely(!dentry)) {
1175 dentry = d_alloc_and_lookup(parent, name, nd);
1176 if (IS_ERR(dentry)) {
1177 mutex_unlock(&dir->i_mutex);
1178 return PTR_ERR(dentry);
1179 }
1180 /* known good */
1181 need_reval = 0;
1182 status = 1;
Josef Bacik44396f42011-05-31 11:58:49 -04001183 } else if (unlikely(d_need_lookup(dentry))) {
1184 dentry = d_inode_lookup(parent, dentry, nd);
1185 if (IS_ERR(dentry)) {
1186 mutex_unlock(&dir->i_mutex);
1187 return PTR_ERR(dentry);
1188 }
1189 /* known good */
1190 need_reval = 0;
1191 status = 1;
Al Viro5a18fff2011-03-11 04:44:53 -05001192 }
1193 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001194 }
Al Viro5a18fff2011-03-11 04:44:53 -05001195 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1196 status = d_revalidate(dentry, nd);
1197 if (unlikely(status <= 0)) {
1198 if (status < 0) {
1199 dput(dentry);
1200 return status;
1201 }
1202 if (!d_invalidate(dentry)) {
1203 dput(dentry);
1204 dentry = NULL;
1205 need_reval = 1;
1206 goto retry;
1207 }
1208 }
1209
David Howells9875cf82011-01-14 18:45:21 +00001210 path->mnt = mnt;
1211 path->dentry = dentry;
1212 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001213 if (unlikely(err < 0)) {
1214 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001215 return err;
Ian Kent89312212011-01-18 12:06:10 +08001216 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001217 if (err)
1218 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001219 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221}
1222
Al Viro52094c82011-02-21 21:34:47 -05001223static inline int may_lookup(struct nameidata *nd)
1224{
1225 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001226 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001227 if (err != -ECHILD)
1228 return err;
Al Viro19660af2011-03-25 10:32:48 -04001229 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001230 return -ECHILD;
1231 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001232 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001233}
1234
Al Viro9856fa12011-03-04 14:22:06 -05001235static inline int handle_dots(struct nameidata *nd, int type)
1236{
1237 if (type == LAST_DOTDOT) {
1238 if (nd->flags & LOOKUP_RCU) {
1239 if (follow_dotdot_rcu(nd))
1240 return -ECHILD;
1241 } else
1242 follow_dotdot(nd);
1243 }
1244 return 0;
1245}
1246
Al Viro951361f2011-03-04 14:44:37 -05001247static void terminate_walk(struct nameidata *nd)
1248{
1249 if (!(nd->flags & LOOKUP_RCU)) {
1250 path_put(&nd->path);
1251 } else {
1252 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001253 if (!(nd->flags & LOOKUP_ROOT))
1254 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001255 rcu_read_unlock();
1256 br_read_unlock(vfsmount_lock);
1257 }
1258}
1259
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001260/*
1261 * Do we need to follow links? We _really_ want to be able
1262 * to do this check without having to look at inode->i_op,
1263 * so we keep a cache of "no, this doesn't need follow_link"
1264 * for the common case.
1265 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001266static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001267{
1268 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1269 if (likely(inode->i_op->follow_link))
1270 return follow;
1271
1272 /* This gets set once for the inode lifetime */
1273 spin_lock(&inode->i_lock);
1274 inode->i_opflags |= IOP_NOFOLLOW;
1275 spin_unlock(&inode->i_lock);
1276 }
1277 return 0;
1278}
1279
Al Viroce57dfc2011-03-13 19:58:58 -04001280static inline int walk_component(struct nameidata *nd, struct path *path,
1281 struct qstr *name, int type, int follow)
1282{
1283 struct inode *inode;
1284 int err;
1285 /*
1286 * "." and ".." are special - ".." especially so because it has
1287 * to be able to know about the current root directory and
1288 * parent relationships.
1289 */
1290 if (unlikely(type != LAST_NORM))
1291 return handle_dots(nd, type);
1292 err = do_lookup(nd, name, path, &inode);
1293 if (unlikely(err)) {
1294 terminate_walk(nd);
1295 return err;
1296 }
1297 if (!inode) {
1298 path_to_nameidata(path, nd);
1299 terminate_walk(nd);
1300 return -ENOENT;
1301 }
Linus Torvalds7813b942011-08-07 09:53:20 -07001302 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001303 if (nd->flags & LOOKUP_RCU) {
1304 if (unlikely(unlazy_walk(nd, path->dentry))) {
1305 terminate_walk(nd);
1306 return -ECHILD;
1307 }
1308 }
Al Viroce57dfc2011-03-13 19:58:58 -04001309 BUG_ON(inode != path->dentry->d_inode);
1310 return 1;
1311 }
1312 path_to_nameidata(path, nd);
1313 nd->inode = inode;
1314 return 0;
1315}
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317/*
Al Virob3563792011-03-14 21:54:55 -04001318 * This limits recursive symlink follows to 8, while
1319 * limiting consecutive symlinks to 40.
1320 *
1321 * Without that kind of total limit, nasty chains of consecutive
1322 * symlinks can cause almost arbitrarily long lookups.
1323 */
1324static inline int nested_symlink(struct path *path, struct nameidata *nd)
1325{
1326 int res;
1327
Al Virob3563792011-03-14 21:54:55 -04001328 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1329 path_put_conditional(path, nd);
1330 path_put(&nd->path);
1331 return -ELOOP;
1332 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001333 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001334
1335 nd->depth++;
1336 current->link_count++;
1337
1338 do {
1339 struct path link = *path;
1340 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001341
1342 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001343 if (!res)
1344 res = walk_component(nd, path, &nd->last,
1345 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001346 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001347 } while (res > 0);
1348
1349 current->link_count--;
1350 nd->depth--;
1351 return res;
1352}
1353
1354/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001355 * We really don't want to look at inode->i_op->lookup
1356 * when we don't have to. So we keep a cache bit in
1357 * the inode ->i_opflags field that says "yes, we can
1358 * do lookup on this inode".
1359 */
1360static inline int can_lookup(struct inode *inode)
1361{
1362 if (likely(inode->i_opflags & IOP_LOOKUP))
1363 return 1;
1364 if (likely(!inode->i_op->lookup))
1365 return 0;
1366
1367 /* We do this once for the lifetime of the inode */
1368 spin_lock(&inode->i_lock);
1369 inode->i_opflags |= IOP_LOOKUP;
1370 spin_unlock(&inode->i_lock);
1371 return 1;
1372}
1373
1374/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001376 * This is the basic name resolution function, turning a pathname into
1377 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001379 * Returns 0 and nd will have valid dentry and mnt on success.
1380 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 */
Al Viro6de88d72009-08-09 01:41:57 +04001382static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
1384 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
1387 while (*name=='/')
1388 name++;
1389 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001390 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 /* At this point we know we have a real path component. */
1393 for(;;) {
1394 unsigned long hash;
1395 struct qstr this;
1396 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001397 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Al Viro52094c82011-02-21 21:34:47 -05001399 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (err)
1401 break;
1402
1403 this.name = name;
1404 c = *(const unsigned char *)name;
1405
1406 hash = init_name_hash();
1407 do {
1408 name++;
1409 hash = partial_name_hash(c, hash);
1410 c = *(const unsigned char *)name;
1411 } while (c && (c != '/'));
1412 this.len = name - (const char *) this.name;
1413 this.hash = end_name_hash(hash);
1414
Al Virofe479a52011-02-22 15:10:03 -05001415 type = LAST_NORM;
1416 if (this.name[0] == '.') switch (this.len) {
1417 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001418 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001419 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001420 nd->flags |= LOOKUP_JUMPED;
1421 }
Al Virofe479a52011-02-22 15:10:03 -05001422 break;
1423 case 1:
1424 type = LAST_DOT;
1425 }
Al Viro5a202bc2011-03-08 14:17:44 -05001426 if (likely(type == LAST_NORM)) {
1427 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001428 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001429 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1430 err = parent->d_op->d_hash(parent, nd->inode,
1431 &this);
1432 if (err < 0)
1433 break;
1434 }
1435 }
Al Virofe479a52011-02-22 15:10:03 -05001436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 /* remove trailing slashes? */
1438 if (!c)
1439 goto last_component;
1440 while (*++name == '/');
1441 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001442 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Al Viroce57dfc2011-03-13 19:58:58 -04001444 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1445 if (err < 0)
1446 return err;
Al Virofe479a52011-02-22 15:10:03 -05001447
Al Viroce57dfc2011-03-13 19:58:58 -04001448 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001449 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001451 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001452 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001453 if (can_lookup(nd->inode))
1454 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001456 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 /* here ends the main loop */
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459last_component:
Al Virob3563792011-03-14 21:54:55 -04001460 nd->last = this;
1461 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001462 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 }
Al Viro951361f2011-03-04 14:44:37 -05001464 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return err;
1466}
1467
Al Viro70e9b352011-03-05 21:12:22 -05001468static int path_init(int dfd, const char *name, unsigned int flags,
1469 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001471 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001472 int fput_needed;
1473 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
1475 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001476 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001478 if (flags & LOOKUP_ROOT) {
1479 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001480 if (*name) {
1481 if (!inode->i_op->lookup)
1482 return -ENOTDIR;
1483 retval = inode_permission(inode, MAY_EXEC);
1484 if (retval)
1485 return retval;
1486 }
Al Viro5b6ca022011-03-09 23:04:47 -05001487 nd->path = nd->root;
1488 nd->inode = inode;
1489 if (flags & LOOKUP_RCU) {
1490 br_read_lock(vfsmount_lock);
1491 rcu_read_lock();
1492 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1493 } else {
1494 path_get(&nd->path);
1495 }
1496 return 0;
1497 }
1498
Al Viro2a737872009-04-07 11:49:53 -04001499 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001502 if (flags & LOOKUP_RCU) {
1503 br_read_lock(vfsmount_lock);
1504 rcu_read_lock();
1505 set_root_rcu(nd);
1506 } else {
1507 set_root(nd);
1508 path_get(&nd->root);
1509 }
Al Viro2a737872009-04-07 11:49:53 -04001510 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001511 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001512 if (flags & LOOKUP_RCU) {
1513 struct fs_struct *fs = current->fs;
1514 unsigned seq;
1515
1516 br_read_lock(vfsmount_lock);
1517 rcu_read_lock();
1518
1519 do {
1520 seq = read_seqcount_begin(&fs->seq);
1521 nd->path = fs->pwd;
1522 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1523 } while (read_seqcount_retry(&fs->seq, seq));
1524 } else {
1525 get_fs_pwd(current->fs, &nd->path);
1526 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001527 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001528 struct dentry *dentry;
1529
Al Viro1abf0c72011-03-13 03:51:11 -04001530 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001531 retval = -EBADF;
1532 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001533 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001534
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001535 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001536
Al Virof52e0c12011-03-14 18:56:51 -04001537 if (*name) {
1538 retval = -ENOTDIR;
1539 if (!S_ISDIR(dentry->d_inode->i_mode))
1540 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001541
Al Viro4ad5abb2011-06-20 19:57:03 -04001542 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001543 if (retval)
1544 goto fput_fail;
1545 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001546
Jan Blunck5dd784d2008-02-14 19:34:38 -08001547 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001548 if (flags & LOOKUP_RCU) {
1549 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001550 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001551 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1552 br_read_lock(vfsmount_lock);
1553 rcu_read_lock();
1554 } else {
1555 path_get(&file->f_path);
1556 fput_light(file, fput_needed);
1557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 }
Al Viroe41f7d42011-02-22 14:02:58 -05001559
Nick Piggin31e6b012011-01-07 17:49:52 +11001560 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001561 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001562
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001563fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001564 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001565out_fail:
1566 return retval;
1567}
1568
Al Virobd92d7f2011-03-14 19:54:59 -04001569static inline int lookup_last(struct nameidata *nd, struct path *path)
1570{
1571 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1572 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1573
1574 nd->flags &= ~LOOKUP_PARENT;
1575 return walk_component(nd, path, &nd->last, nd->last_type,
1576 nd->flags & LOOKUP_FOLLOW);
1577}
1578
Al Viro9b4a9b12009-04-07 11:44:16 -04001579/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001580static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001581 unsigned int flags, struct nameidata *nd)
1582{
Al Viro70e9b352011-03-05 21:12:22 -05001583 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001584 struct path path;
1585 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001586
1587 /*
1588 * Path walking is largely split up into 2 different synchronisation
1589 * schemes, rcu-walk and ref-walk (explained in
1590 * Documentation/filesystems/path-lookup.txt). These share much of the
1591 * path walk code, but some things particularly setup, cleanup, and
1592 * following mounts are sufficiently divergent that functions are
1593 * duplicated. Typically there is a function foo(), and its RCU
1594 * analogue, foo_rcu().
1595 *
1596 * -ECHILD is the error number of choice (just to avoid clashes) that
1597 * is returned if some aspect of an rcu-walk fails. Such an error must
1598 * be handled by restarting a traditional ref-walk (which will always
1599 * be able to complete).
1600 */
Al Virobd92d7f2011-03-14 19:54:59 -04001601 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001602
Al Virobd92d7f2011-03-14 19:54:59 -04001603 if (unlikely(err))
1604 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001605
1606 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001607 err = link_path_walk(name, nd);
1608
1609 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001610 err = lookup_last(nd, &path);
1611 while (err > 0) {
1612 void *cookie;
1613 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001614 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001615 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001616 if (!err)
1617 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001618 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001619 }
1620 }
Al Viroee0827c2011-02-21 23:38:09 -05001621
Al Viro9f1fafe2011-03-25 11:00:12 -04001622 if (!err)
1623 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001624
1625 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1626 if (!nd->inode->i_op->lookup) {
1627 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001628 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001629 }
1630 }
Al Viro16c2cd72011-02-22 15:50:10 -05001631
Al Viro70e9b352011-03-05 21:12:22 -05001632 if (base)
1633 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001634
Al Viro5b6ca022011-03-09 23:04:47 -05001635 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001636 path_put(&nd->root);
1637 nd->root.mnt = NULL;
1638 }
Al Virobd92d7f2011-03-14 19:54:59 -04001639 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001640}
Nick Piggin31e6b012011-01-07 17:49:52 +11001641
Al Viroee0827c2011-02-21 23:38:09 -05001642static int do_path_lookup(int dfd, const char *name,
1643 unsigned int flags, struct nameidata *nd)
1644{
1645 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1646 if (unlikely(retval == -ECHILD))
1647 retval = path_lookupat(dfd, name, flags, nd);
1648 if (unlikely(retval == -ESTALE))
1649 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001650
1651 if (likely(!retval)) {
1652 if (unlikely(!audit_dummy_context())) {
1653 if (nd->path.dentry && nd->inode)
1654 audit_inode(name, nd->path.dentry);
1655 }
1656 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001657 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658}
1659
Al Viroc9c6cac2011-02-16 15:15:47 -05001660int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001661{
Al Viroc9c6cac2011-02-16 15:15:47 -05001662 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001663}
1664
Al Virod1811462008-08-02 00:49:18 -04001665int kern_path(const char *name, unsigned int flags, struct path *path)
1666{
1667 struct nameidata nd;
1668 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1669 if (!res)
1670 *path = nd.path;
1671 return res;
1672}
1673
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001674/**
1675 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1676 * @dentry: pointer to dentry of the base directory
1677 * @mnt: pointer to vfs mount of the base directory
1678 * @name: pointer to file name
1679 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001680 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001681 */
1682int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1683 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001684 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001685{
Al Viroe0a01242011-06-27 17:00:37 -04001686 struct nameidata nd;
1687 int err;
1688 nd.root.dentry = dentry;
1689 nd.root.mnt = mnt;
1690 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001691 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001692 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1693 if (!err)
1694 *path = nd.path;
1695 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001696}
1697
Christoph Hellwigeead1912007-10-16 23:25:38 -07001698static struct dentry *__lookup_hash(struct qstr *name,
1699 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001701 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001702 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 int err;
1704
Al Viro4ad5abb2011-06-20 19:57:03 -04001705 err = inode_permission(inode, MAY_EXEC);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001706 if (err)
1707 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001710 * Don't bother with __d_lookup: callers are for creat as
1711 * well as unlink, so a lot of the time it would cost
1712 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001713 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001714 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001715
Josef Bacik44396f42011-05-31 11:58:49 -04001716 if (dentry && d_need_lookup(dentry)) {
1717 /*
1718 * __lookup_hash is called with the parent dir's i_mutex already
1719 * held, so we are good to go here.
1720 */
1721 dentry = d_inode_lookup(base, dentry, nd);
1722 if (IS_ERR(dentry))
1723 return dentry;
1724 }
1725
Al Virod2d9e9f2011-06-20 10:55:26 -04001726 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1727 int status = d_revalidate(dentry, nd);
1728 if (unlikely(status <= 0)) {
1729 /*
1730 * The dentry failed validation.
1731 * If d_revalidate returned 0 attempt to invalidate
1732 * the dentry otherwise d_revalidate is asking us
1733 * to return a fail status.
1734 */
1735 if (status < 0) {
1736 dput(dentry);
1737 return ERR_PTR(status);
1738 } else if (!d_invalidate(dentry)) {
1739 dput(dentry);
1740 dentry = NULL;
1741 }
1742 }
1743 }
Al Viro6e6b1bd2009-08-13 23:38:37 +04001744
Nick Pigginbaa03892010-08-18 04:37:31 +10001745 if (!dentry)
1746 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 return dentry;
1749}
1750
James Morris057f6c02007-04-26 00:12:05 -07001751/*
1752 * Restricted form of lookup. Doesn't follow links, single-component only,
1753 * needs parent already locked. Doesn't follow mounts.
1754 * SMP-safe.
1755 */
Adrian Bunka244e162006-03-31 02:32:11 -08001756static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757{
Jan Blunck4ac91372008-02-14 19:34:32 -08001758 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759}
1760
Christoph Hellwigeead1912007-10-16 23:25:38 -07001761/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001762 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001763 * @name: pathname component to lookup
1764 * @base: base directory to lookup from
1765 * @len: maximum length @len should be interpreted to
1766 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001767 * Note that this routine is purely a helper for filesystem usage and should
1768 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001769 * nameidata argument is passed to the filesystem methods and a filesystem
1770 * using this helper needs to be prepared for that.
1771 */
James Morris057f6c02007-04-26 00:12:05 -07001772struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1773{
James Morris057f6c02007-04-26 00:12:05 -07001774 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001775 unsigned long hash;
1776 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001777
David Woodhouse2f9092e2009-04-20 23:18:37 +01001778 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1779
Al Viro6a96ba52011-03-07 23:49:20 -05001780 this.name = name;
1781 this.len = len;
1782 if (!len)
1783 return ERR_PTR(-EACCES);
1784
1785 hash = init_name_hash();
1786 while (len--) {
1787 c = *(const unsigned char *)name++;
1788 if (c == '/' || c == '\0')
1789 return ERR_PTR(-EACCES);
1790 hash = partial_name_hash(c, hash);
1791 }
1792 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001793 /*
1794 * See if the low-level filesystem might want
1795 * to use its own hash..
1796 */
1797 if (base->d_flags & DCACHE_OP_HASH) {
1798 int err = base->d_op->d_hash(base, base->d_inode, &this);
1799 if (err < 0)
1800 return ERR_PTR(err);
1801 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001802
Christoph Hellwig49705b72005-11-08 21:35:06 -08001803 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001804}
1805
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001806int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1807 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808{
Al Viro2d8f3032008-07-22 09:59:21 -04001809 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001810 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001813
1814 BUG_ON(flags & LOOKUP_PARENT);
1815
1816 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001818 if (!err)
1819 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821 return err;
1822}
1823
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001824int user_path_at(int dfd, const char __user *name, unsigned flags,
1825 struct path *path)
1826{
1827 return user_path_at_empty(dfd, name, flags, path, 0);
1828}
1829
Al Viro2ad94ae2008-07-21 09:32:51 -04001830static int user_path_parent(int dfd, const char __user *path,
1831 struct nameidata *nd, char **name)
1832{
1833 char *s = getname(path);
1834 int error;
1835
1836 if (IS_ERR(s))
1837 return PTR_ERR(s);
1838
1839 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1840 if (error)
1841 putname(s);
1842 else
1843 *name = s;
1844
1845 return error;
1846}
1847
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848/*
1849 * It's inline, so penalty for filesystems that don't use sticky bit is
1850 * minimal.
1851 */
1852static inline int check_sticky(struct inode *dir, struct inode *inode)
1853{
David Howellsda9592e2008-11-14 10:39:05 +11001854 uid_t fsuid = current_fsuid();
1855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 if (!(dir->i_mode & S_ISVTX))
1857 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001858 if (current_user_ns() != inode_userns(inode))
1859 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001860 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001862 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001864
1865other_userns:
1866 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867}
1868
1869/*
1870 * Check whether we can remove a link victim from directory dir, check
1871 * whether the type of victim is right.
1872 * 1. We can't do it if dir is read-only (done in permission())
1873 * 2. We should have write and exec permissions on dir
1874 * 3. We can't remove anything from append-only dir
1875 * 4. We can't do anything with immutable dir (done in permission())
1876 * 5. If the sticky bit on dir is set we should either
1877 * a. be owner of dir, or
1878 * b. be owner of victim, or
1879 * c. have CAP_FOWNER capability
1880 * 6. If the victim is append-only or immutable we can't do antyhing with
1881 * links pointing to it.
1882 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1883 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1884 * 9. We can't remove a root or mountpoint.
1885 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1886 * nfs_async_unlink().
1887 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001888static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
1890 int error;
1891
1892 if (!victim->d_inode)
1893 return -ENOENT;
1894
1895 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001896 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Al Virof419a2e2008-07-22 00:07:17 -04001898 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (error)
1900 return error;
1901 if (IS_APPEND(dir))
1902 return -EPERM;
1903 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001904 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 return -EPERM;
1906 if (isdir) {
1907 if (!S_ISDIR(victim->d_inode->i_mode))
1908 return -ENOTDIR;
1909 if (IS_ROOT(victim))
1910 return -EBUSY;
1911 } else if (S_ISDIR(victim->d_inode->i_mode))
1912 return -EISDIR;
1913 if (IS_DEADDIR(dir))
1914 return -ENOENT;
1915 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1916 return -EBUSY;
1917 return 0;
1918}
1919
1920/* Check whether we can create an object with dentry child in directory
1921 * dir.
1922 * 1. We can't do it if child already exists (open has special treatment for
1923 * this case, but since we are inlined it's OK)
1924 * 2. We can't do it if dir is read-only (done in permission())
1925 * 3. We should have write and exec permissions on dir
1926 * 4. We can't do it if dir is immutable (done in permission())
1927 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001928static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929{
1930 if (child->d_inode)
1931 return -EEXIST;
1932 if (IS_DEADDIR(dir))
1933 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001934 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935}
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937/*
1938 * p1 and p2 should be directories on the same fs.
1939 */
1940struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1941{
1942 struct dentry *p;
1943
1944 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001945 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return NULL;
1947 }
1948
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001949 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001951 p = d_ancestor(p2, p1);
1952 if (p) {
1953 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1954 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1955 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 }
1957
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001958 p = d_ancestor(p1, p2);
1959 if (p) {
1960 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1961 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1962 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 }
1964
Ingo Molnarf2eace22006-07-03 00:25:05 -07001965 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1966 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 return NULL;
1968}
1969
1970void unlock_rename(struct dentry *p1, struct dentry *p2)
1971{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001972 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001974 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001975 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 }
1977}
1978
1979int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1980 struct nameidata *nd)
1981{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001982 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
1984 if (error)
1985 return error;
1986
Al Viroacfa4382008-12-04 10:06:33 -05001987 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 return -EACCES; /* shouldn't it be ENOSYS? */
1989 mode &= S_IALLUGO;
1990 mode |= S_IFREG;
1991 error = security_inode_create(dir, dentry, mode);
1992 if (error)
1993 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001995 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001996 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return error;
1998}
1999
Al Viro73d049a2011-03-11 12:08:24 -05002000static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002002 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 struct inode *inode = dentry->d_inode;
2004 int error;
2005
Al Virobcda7652011-03-13 16:42:14 -04002006 /* O_PATH? */
2007 if (!acc_mode)
2008 return 0;
2009
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 if (!inode)
2011 return -ENOENT;
2012
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002013 switch (inode->i_mode & S_IFMT) {
2014 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002016 case S_IFDIR:
2017 if (acc_mode & MAY_WRITE)
2018 return -EISDIR;
2019 break;
2020 case S_IFBLK:
2021 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002022 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002024 /*FALLTHRU*/
2025 case S_IFIFO:
2026 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002028 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002029 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002030
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002031 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002032 if (error)
2033 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 /*
2036 * An append-only file must be opened in append mode for writing.
2037 */
2038 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002039 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002040 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002042 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
2044
2045 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002046 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002047 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002049 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002050}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
Jeff Laytone1181ee2010-12-07 16:19:50 -05002052static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002053{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002054 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002055 struct inode *inode = path->dentry->d_inode;
2056 int error = get_write_access(inode);
2057 if (error)
2058 return error;
2059 /*
2060 * Refuse to truncate files with mandatory locks held on them.
2061 */
2062 error = locks_verify_locked(inode);
2063 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002064 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002065 if (!error) {
2066 error = do_truncate(path->dentry, 0,
2067 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002068 filp);
Al Viro7715b522009-12-16 03:54:00 -05002069 }
2070 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002071 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072}
2073
Dave Hansend57999e2008-02-15 14:37:27 -08002074static inline int open_to_namei_flags(int flag)
2075{
Al Viro8a5e9292011-06-25 19:15:54 -04002076 if ((flag & O_ACCMODE) == 3)
2077 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002078 return flag;
2079}
2080
Nick Piggin31e6b012011-01-07 17:49:52 +11002081/*
Al Virofe2d35f2011-03-05 22:58:25 -05002082 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002083 */
Al Virofb1cc552009-12-24 01:58:28 -05002084static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002085 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002086{
Al Viroa1e28032009-12-24 02:12:06 -05002087 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002088 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002089 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002090 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002091 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002092 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002093 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002094 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002095
Al Viroc3e380b2011-02-23 13:39:45 -05002096 nd->flags &= ~LOOKUP_PARENT;
2097 nd->flags |= op->intent;
2098
Al Viro1f36f772009-12-26 10:56:19 -05002099 switch (nd->last_type) {
2100 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002101 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002102 error = handle_dots(nd, nd->last_type);
2103 if (error)
2104 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002105 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002106 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002107 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002108 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002109 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002110 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002111 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002112 error = -EISDIR;
2113 goto exit;
2114 }
2115 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002116 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002117 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002118 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002119 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002120 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002121 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002122 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002123
Al Viroca344a892011-03-09 00:36:45 -05002124 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002125 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002126 if (nd->last.name[nd->last.len])
2127 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002128 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2129 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002130 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002131 error = walk_component(nd, path, &nd->last, LAST_NORM,
2132 !symlink_ok);
2133 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002134 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002135 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002136 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002137 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002138 error = complete_walk(nd);
2139 if (error)
2140 return ERR_PTR(-ECHILD);
Al Virofe2d35f2011-03-05 22:58:25 -05002141
2142 error = -ENOTDIR;
2143 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002144 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002145 goto exit;
2146 }
2147 audit_inode(pathname, nd->path.dentry);
2148 goto ok;
2149 }
2150
2151 /* create side of things */
Al Viroa3fbbde2011-11-07 21:21:26 +00002152 /*
2153 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2154 * cleared when we got to the last component we are about to look up
2155 */
Al Viro9f1fafe2011-03-25 11:00:12 -04002156 error = complete_walk(nd);
2157 if (error)
2158 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002159
2160 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002161 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002162 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002163 if (nd->last.name[nd->last.len])
2164 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002165
Al Viroa1e28032009-12-24 02:12:06 -05002166 mutex_lock(&dir->d_inode->i_mutex);
2167
Al Viro6c0d46c2011-03-09 00:59:59 -05002168 dentry = lookup_hash(nd);
2169 error = PTR_ERR(dentry);
2170 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002171 mutex_unlock(&dir->d_inode->i_mutex);
2172 goto exit;
2173 }
2174
Al Viro6c0d46c2011-03-09 00:59:59 -05002175 path->dentry = dentry;
2176 path->mnt = nd->path.mnt;
2177
Al Virofb1cc552009-12-24 01:58:28 -05002178 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002179 if (!dentry->d_inode) {
2180 int mode = op->mode;
2181 if (!IS_POSIXACL(dir->d_inode))
2182 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002183 /*
2184 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002185 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002186 * the time when the file is created and when
2187 * a permanent write count is taken through
2188 * the 'struct file' in nameidata_to_filp().
2189 */
2190 error = mnt_want_write(nd->path.mnt);
2191 if (error)
2192 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002193 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002194 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002195 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002196 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002197 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002198 error = security_path_mknod(&nd->path, dentry, mode, 0);
2199 if (error)
2200 goto exit_mutex_unlock;
2201 error = vfs_create(dir->d_inode, dentry, mode, nd);
2202 if (error)
2203 goto exit_mutex_unlock;
2204 mutex_unlock(&dir->d_inode->i_mutex);
2205 dput(nd->path.dentry);
2206 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002207 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002208 }
2209
2210 /*
2211 * It already exists.
2212 */
2213 mutex_unlock(&dir->d_inode->i_mutex);
2214 audit_inode(pathname, path->dentry);
2215
2216 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002217 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002218 goto exit_dput;
2219
David Howells9875cf82011-01-14 18:45:21 +00002220 error = follow_managed(path, nd->flags);
2221 if (error < 0)
2222 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002223
Al Viroa3fbbde2011-11-07 21:21:26 +00002224 if (error)
2225 nd->flags |= LOOKUP_JUMPED;
2226
Al Virofb1cc552009-12-24 01:58:28 -05002227 error = -ENOENT;
2228 if (!path->dentry->d_inode)
2229 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002230
2231 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002232 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002233
2234 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002235 nd->inode = path->dentry->d_inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002236 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2237 error = complete_walk(nd);
2238 if (error)
2239 goto exit;
Al Virofb1cc552009-12-24 01:58:28 -05002240 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002241 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002242 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002243ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002244 if (!S_ISREG(nd->inode->i_mode))
2245 will_truncate = 0;
2246
Al Viro0f9d1a12011-03-09 00:13:14 -05002247 if (will_truncate) {
2248 error = mnt_want_write(nd->path.mnt);
2249 if (error)
2250 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002251 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002252 }
Al Viroca344a892011-03-09 00:36:45 -05002253common:
Al Virobcda7652011-03-13 16:42:14 -04002254 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002255 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002256 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002257 filp = nameidata_to_filp(nd);
2258 if (!IS_ERR(filp)) {
2259 error = ima_file_check(filp, op->acc_mode);
2260 if (error) {
2261 fput(filp);
2262 filp = ERR_PTR(error);
2263 }
2264 }
2265 if (!IS_ERR(filp)) {
2266 if (will_truncate) {
2267 error = handle_truncate(filp);
2268 if (error) {
2269 fput(filp);
2270 filp = ERR_PTR(error);
2271 }
2272 }
2273 }
Al Viroca344a892011-03-09 00:36:45 -05002274out:
2275 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002276 mnt_drop_write(nd->path.mnt);
2277 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002278 return filp;
2279
2280exit_mutex_unlock:
2281 mutex_unlock(&dir->d_inode->i_mutex);
2282exit_dput:
2283 path_put_conditional(path, nd);
2284exit:
Al Viroca344a892011-03-09 00:36:45 -05002285 filp = ERR_PTR(error);
2286 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002287}
2288
Al Viro13aab422011-02-23 17:54:08 -05002289static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002290 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
Al Virofe2d35f2011-03-05 22:58:25 -05002292 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002293 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002294 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002295 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002296
2297 filp = get_empty_filp();
2298 if (!filp)
2299 return ERR_PTR(-ENFILE);
2300
Al Viro47c805d2011-02-23 17:44:09 -05002301 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002302 nd->intent.open.file = filp;
2303 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2304 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002305
Al Viro73d049a2011-03-11 12:08:24 -05002306 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002307 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002308 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002309
Al Virofe2d35f2011-03-05 22:58:25 -05002310 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002311 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002312 if (unlikely(error))
2313 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
Al Viro73d049a2011-03-11 12:08:24 -05002315 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002316 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002317 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002318 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002319 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002320 path_put_conditional(&path, nd);
2321 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002322 filp = ERR_PTR(-ELOOP);
2323 break;
2324 }
Al Viro73d049a2011-03-11 12:08:24 -05002325 nd->flags |= LOOKUP_PARENT;
2326 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002327 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002328 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002329 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002330 else
Al Viro73d049a2011-03-11 12:08:24 -05002331 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002332 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002333 }
Al Viro10fa8e62009-12-26 07:09:49 -05002334out:
Al Viro73d049a2011-03-11 12:08:24 -05002335 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2336 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002337 if (base)
2338 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002339 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002340 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Nick Piggin31e6b012011-01-07 17:49:52 +11002342out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002343 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002344 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345}
2346
Al Viro13aab422011-02-23 17:54:08 -05002347struct file *do_filp_open(int dfd, const char *pathname,
2348 const struct open_flags *op, int flags)
2349{
Al Viro73d049a2011-03-11 12:08:24 -05002350 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002351 struct file *filp;
2352
Al Viro73d049a2011-03-11 12:08:24 -05002353 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002354 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002355 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002356 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002357 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002358 return filp;
2359}
2360
Al Viro73d049a2011-03-11 12:08:24 -05002361struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2362 const char *name, const struct open_flags *op, int flags)
2363{
2364 struct nameidata nd;
2365 struct file *file;
2366
2367 nd.root.mnt = mnt;
2368 nd.root.dentry = dentry;
2369
2370 flags |= LOOKUP_ROOT;
2371
Al Virobcda7652011-03-13 16:42:14 -04002372 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002373 return ERR_PTR(-ELOOP);
2374
2375 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2376 if (unlikely(file == ERR_PTR(-ECHILD)))
2377 file = path_openat(-1, name, &nd, op, flags);
2378 if (unlikely(file == ERR_PTR(-ESTALE)))
2379 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2380 return file;
2381}
2382
Al Viroed75e952011-06-27 16:53:43 -04002383struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002385 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002386 struct nameidata nd;
2387 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2388 if (error)
2389 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002391 /*
2392 * Yucky last component or no last component at all?
2393 * (foo/., foo/.., /////)
2394 */
Al Viroed75e952011-06-27 16:53:43 -04002395 if (nd.last_type != LAST_NORM)
2396 goto out;
2397 nd.flags &= ~LOOKUP_PARENT;
2398 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2399 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002400
2401 /*
2402 * Do the final lookup.
2403 */
Al Viroed75e952011-06-27 16:53:43 -04002404 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2405 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 if (IS_ERR(dentry))
2407 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002408
Al Viroe9baf6e2008-05-15 04:49:12 -04002409 if (dentry->d_inode)
2410 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002411 /*
2412 * Special case - lookup gave negative, but... we had foo/bar/
2413 * From the vfs_mknod() POV we just have a negative dentry -
2414 * all is fine. Let's be bastards - you had / on the end, you've
2415 * been asking for (non-existent) directory. -ENOENT for you.
2416 */
Al Viroed75e952011-06-27 16:53:43 -04002417 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002418 dput(dentry);
2419 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002420 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002421 }
Al Viroed75e952011-06-27 16:53:43 -04002422 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002424eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002426 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427fail:
Al Viroed75e952011-06-27 16:53:43 -04002428 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2429out:
2430 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 return dentry;
2432}
Al Virodae6ad82011-06-26 11:50:15 -04002433EXPORT_SYMBOL(kern_path_create);
2434
2435struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2436{
2437 char *tmp = getname(pathname);
2438 struct dentry *res;
2439 if (IS_ERR(tmp))
2440 return ERR_CAST(tmp);
2441 res = kern_path_create(dfd, tmp, path, is_dir);
2442 putname(tmp);
2443 return res;
2444}
2445EXPORT_SYMBOL(user_path_create);
2446
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2448{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002449 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
2451 if (error)
2452 return error;
2453
Serge E. Hallyne795b712011-03-23 16:43:25 -07002454 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2455 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 return -EPERM;
2457
Al Viroacfa4382008-12-04 10:06:33 -05002458 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 return -EPERM;
2460
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002461 error = devcgroup_inode_mknod(mode, dev);
2462 if (error)
2463 return error;
2464
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 error = security_inode_mknod(dir, dentry, mode, dev);
2466 if (error)
2467 return error;
2468
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002470 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002471 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 return error;
2473}
2474
Dave Hansen463c3192008-02-15 14:37:57 -08002475static int may_mknod(mode_t mode)
2476{
2477 switch (mode & S_IFMT) {
2478 case S_IFREG:
2479 case S_IFCHR:
2480 case S_IFBLK:
2481 case S_IFIFO:
2482 case S_IFSOCK:
2483 case 0: /* zero mode translates to S_IFREG */
2484 return 0;
2485 case S_IFDIR:
2486 return -EPERM;
2487 default:
2488 return -EINVAL;
2489 }
2490}
2491
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002492SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2493 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494{
Al Viro2ad94ae2008-07-21 09:32:51 -04002495 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002496 struct path path;
2497 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
2499 if (S_ISDIR(mode))
2500 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501
Al Virodae6ad82011-06-26 11:50:15 -04002502 dentry = user_path_create(dfd, filename, &path, 0);
2503 if (IS_ERR(dentry))
2504 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002505
Al Virodae6ad82011-06-26 11:50:15 -04002506 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002507 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002508 error = may_mknod(mode);
2509 if (error)
2510 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002511 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002512 if (error)
2513 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002514 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002515 if (error)
2516 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002517 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002519 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 break;
2521 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002522 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 new_decode_dev(dev));
2524 break;
2525 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002526 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002529out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002530 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002531out_dput:
2532 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002533 mutex_unlock(&path.dentry->d_inode->i_mutex);
2534 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
2536 return error;
2537}
2538
Heiko Carstens3480b252009-01-14 14:14:16 +01002539SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002540{
2541 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2542}
2543
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2545{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002546 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
2548 if (error)
2549 return error;
2550
Al Viroacfa4382008-12-04 10:06:33 -05002551 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return -EPERM;
2553
2554 mode &= (S_IRWXUGO|S_ISVTX);
2555 error = security_inode_mkdir(dir, dentry, mode);
2556 if (error)
2557 return error;
2558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002560 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002561 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 return error;
2563}
2564
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002565SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566{
Dave Hansen6902d922006-09-30 23:29:01 -07002567 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002568 struct path path;
2569 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Al Virodae6ad82011-06-26 11:50:15 -04002571 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002572 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002573 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002574
Al Virodae6ad82011-06-26 11:50:15 -04002575 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002576 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002577 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002578 if (error)
2579 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002580 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002581 if (error)
2582 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002583 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002584out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002585 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002586out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002587 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002588 mutex_unlock(&path.dentry->d_inode->i_mutex);
2589 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 return error;
2591}
2592
Heiko Carstens3cdad422009-01-14 14:14:22 +01002593SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002594{
2595 return sys_mkdirat(AT_FDCWD, pathname, mode);
2596}
2597
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598/*
Sage Weila71905f2011-05-24 13:06:08 -07002599 * The dentry_unhash() helper will try to drop the dentry early: we
2600 * should have a usage count of 2 if we're the only user of this
2601 * dentry, and if that is true (possibly after pruning the dcache),
2602 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 *
2604 * A low-level filesystem can, if it choses, legally
2605 * do a
2606 *
2607 * if (!d_unhashed(dentry))
2608 * return -EBUSY;
2609 *
2610 * if it cannot handle the case of removing a directory
2611 * that is still in use by something else..
2612 */
2613void dentry_unhash(struct dentry *dentry)
2614{
Vasily Averindc168422006-12-06 20:37:07 -08002615 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002617 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 __d_drop(dentry);
2619 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620}
2621
2622int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2623{
2624 int error = may_delete(dir, dentry, 1);
2625
2626 if (error)
2627 return error;
2628
Al Viroacfa4382008-12-04 10:06:33 -05002629 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return -EPERM;
2631
Al Viro1d2ef592011-09-14 18:55:41 +01002632 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002633 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
Sage Weil912dbc12011-05-24 13:06:11 -07002635 error = -EBUSY;
2636 if (d_mountpoint(dentry))
2637 goto out;
2638
2639 error = security_inode_rmdir(dir, dentry);
2640 if (error)
2641 goto out;
2642
Sage Weil3cebde22011-05-29 21:20:59 -07002643 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002644 error = dir->i_op->rmdir(dir, dentry);
2645 if (error)
2646 goto out;
2647
2648 dentry->d_inode->i_flags |= S_DEAD;
2649 dont_mount(dentry);
2650
2651out:
2652 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01002653 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002654 if (!error)
2655 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 return error;
2657}
2658
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002659static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660{
2661 int error = 0;
2662 char * name;
2663 struct dentry *dentry;
2664 struct nameidata nd;
2665
Al Viro2ad94ae2008-07-21 09:32:51 -04002666 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002668 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
2670 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002671 case LAST_DOTDOT:
2672 error = -ENOTEMPTY;
2673 goto exit1;
2674 case LAST_DOT:
2675 error = -EINVAL;
2676 goto exit1;
2677 case LAST_ROOT:
2678 error = -EBUSY;
2679 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002681
2682 nd.flags &= ~LOOKUP_PARENT;
2683
Jan Blunck4ac91372008-02-14 19:34:32 -08002684 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002685 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002687 if (IS_ERR(dentry))
2688 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002689 if (!dentry->d_inode) {
2690 error = -ENOENT;
2691 goto exit3;
2692 }
Dave Hansen06227532008-02-15 14:37:34 -08002693 error = mnt_want_write(nd.path.mnt);
2694 if (error)
2695 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002696 error = security_path_rmdir(&nd.path, dentry);
2697 if (error)
2698 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002699 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002700exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002701 mnt_drop_write(nd.path.mnt);
2702exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002703 dput(dentry);
2704exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002705 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002707 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 putname(name);
2709 return error;
2710}
2711
Heiko Carstens3cdad422009-01-14 14:14:22 +01002712SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002713{
2714 return do_rmdir(AT_FDCWD, pathname);
2715}
2716
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717int vfs_unlink(struct inode *dir, struct dentry *dentry)
2718{
2719 int error = may_delete(dir, dentry, 0);
2720
2721 if (error)
2722 return error;
2723
Al Viroacfa4382008-12-04 10:06:33 -05002724 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 return -EPERM;
2726
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002727 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 if (d_mountpoint(dentry))
2729 error = -EBUSY;
2730 else {
2731 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002732 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002734 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002735 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002738 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739
2740 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2741 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002742 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 }
Robert Love0eeca282005-07-12 17:06:03 -04002745
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 return error;
2747}
2748
2749/*
2750 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002751 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 * writeout happening, and we don't want to prevent access to the directory
2753 * while waiting on the I/O.
2754 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002755static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756{
Al Viro2ad94ae2008-07-21 09:32:51 -04002757 int error;
2758 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 struct dentry *dentry;
2760 struct nameidata nd;
2761 struct inode *inode = NULL;
2762
Al Viro2ad94ae2008-07-21 09:32:51 -04002763 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002765 return error;
2766
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 error = -EISDIR;
2768 if (nd.last_type != LAST_NORM)
2769 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002770
2771 nd.flags &= ~LOOKUP_PARENT;
2772
Jan Blunck4ac91372008-02-14 19:34:32 -08002773 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002774 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 error = PTR_ERR(dentry);
2776 if (!IS_ERR(dentry)) {
2777 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002778 if (nd.last.name[nd.last.len])
2779 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002781 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002782 goto slashes;
2783 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002784 error = mnt_want_write(nd.path.mnt);
2785 if (error)
2786 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002787 error = security_path_unlink(&nd.path, dentry);
2788 if (error)
2789 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002790 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002791exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002792 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 exit2:
2794 dput(dentry);
2795 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002796 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (inode)
2798 iput(inode); /* truncate the inode here */
2799exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002800 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 putname(name);
2802 return error;
2803
2804slashes:
2805 error = !dentry->d_inode ? -ENOENT :
2806 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2807 goto exit2;
2808}
2809
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002810SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002811{
2812 if ((flag & ~AT_REMOVEDIR) != 0)
2813 return -EINVAL;
2814
2815 if (flag & AT_REMOVEDIR)
2816 return do_rmdir(dfd, pathname);
2817
2818 return do_unlinkat(dfd, pathname);
2819}
2820
Heiko Carstens3480b252009-01-14 14:14:16 +01002821SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002822{
2823 return do_unlinkat(AT_FDCWD, pathname);
2824}
2825
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002826int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002828 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
2830 if (error)
2831 return error;
2832
Al Viroacfa4382008-12-04 10:06:33 -05002833 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 return -EPERM;
2835
2836 error = security_inode_symlink(dir, dentry, oldname);
2837 if (error)
2838 return error;
2839
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002841 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002842 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 return error;
2844}
2845
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002846SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2847 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848{
Al Viro2ad94ae2008-07-21 09:32:51 -04002849 int error;
2850 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07002851 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002852 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853
2854 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002855 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002857
Al Virodae6ad82011-06-26 11:50:15 -04002858 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07002859 error = PTR_ERR(dentry);
2860 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002861 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07002862
Al Virodae6ad82011-06-26 11:50:15 -04002863 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002864 if (error)
2865 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002866 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002867 if (error)
2868 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002869 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002870out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002871 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002872out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002873 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002874 mutex_unlock(&path.dentry->d_inode->i_mutex);
2875 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07002876out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 putname(from);
2878 return error;
2879}
2880
Heiko Carstens3480b252009-01-14 14:14:16 +01002881SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002882{
2883 return sys_symlinkat(oldname, AT_FDCWD, newname);
2884}
2885
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2887{
2888 struct inode *inode = old_dentry->d_inode;
2889 int error;
2890
2891 if (!inode)
2892 return -ENOENT;
2893
Miklos Szeredia95164d2008-07-30 15:08:48 +02002894 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 if (error)
2896 return error;
2897
2898 if (dir->i_sb != inode->i_sb)
2899 return -EXDEV;
2900
2901 /*
2902 * A link to an append-only or immutable file cannot be created.
2903 */
2904 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2905 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002906 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002908 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 return -EPERM;
2910
2911 error = security_inode_link(old_dentry, dir, new_dentry);
2912 if (error)
2913 return error;
2914
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002915 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302916 /* Make sure we don't allow creating hardlink to an unlinked file */
2917 if (inode->i_nlink == 0)
2918 error = -ENOENT;
2919 else
2920 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002921 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002922 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002923 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 return error;
2925}
2926
2927/*
2928 * Hardlinks are often used in delicate situations. We avoid
2929 * security-related surprises by not following symlinks on the
2930 * newname. --KAB
2931 *
2932 * We don't follow them on the oldname either to be compatible
2933 * with linux 2.0, and to avoid hard-linking to directories
2934 * and other special files. --ADM
2935 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002936SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2937 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938{
2939 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002940 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302941 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302944 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002945 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302946 /*
2947 * To use null names we require CAP_DAC_READ_SEARCH
2948 * This ensures that not everyone will be able to create
2949 * handlink using the passed filedescriptor.
2950 */
2951 if (flags & AT_EMPTY_PATH) {
2952 if (!capable(CAP_DAC_READ_SEARCH))
2953 return -ENOENT;
2954 how = LOOKUP_EMPTY;
2955 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002956
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302957 if (flags & AT_SYMLINK_FOLLOW)
2958 how |= LOOKUP_FOLLOW;
2959
2960 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002962 return error;
2963
Al Virodae6ad82011-06-26 11:50:15 -04002964 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002966 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002967 goto out;
2968
2969 error = -EXDEV;
2970 if (old_path.mnt != new_path.mnt)
2971 goto out_dput;
2972 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002973 if (error)
2974 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002975 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002976 if (error)
2977 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002978 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002979out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002980 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002981out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002982 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002983 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2984 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985out:
Al Viro2d8f3032008-07-22 09:59:21 -04002986 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
2988 return error;
2989}
2990
Heiko Carstens3480b252009-01-14 14:14:16 +01002991SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002992{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002993 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002994}
2995
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996/*
2997 * The worst of all namespace operations - renaming directory. "Perverted"
2998 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2999 * Problems:
3000 * a) we can get into loop creation. Check is done in is_subdir().
3001 * b) race potential - two innocent renames can create a loop together.
3002 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003003 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 * story.
3005 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003006 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 * whether the target exists). Solution: try to be smart with locking
3008 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003009 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 * move will be locked. Thus we can rank directories by the tree
3011 * (ancestors first) and rank all non-directories after them.
3012 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003013 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 * HOWEVER, it relies on the assumption that any object with ->lookup()
3015 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3016 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003017 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003018 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003020 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 * locking].
3022 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003023static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3024 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025{
3026 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003027 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
3029 /*
3030 * If we are going to change the parent - check write permissions,
3031 * we'll need to flip '..'.
3032 */
3033 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003034 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 if (error)
3036 return error;
3037 }
3038
3039 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3040 if (error)
3041 return error;
3042
Al Viro1d2ef592011-09-14 18:55:41 +01003043 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003044 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003045 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003046
3047 error = -EBUSY;
3048 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3049 goto out;
3050
Sage Weil3cebde22011-05-29 21:20:59 -07003051 if (target)
3052 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003053 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3054 if (error)
3055 goto out;
3056
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003058 target->i_flags |= S_DEAD;
3059 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 }
Sage Weil9055cba2011-05-24 13:06:12 -07003061out:
3062 if (target)
3063 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003064 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003065 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003066 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3067 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 return error;
3069}
3070
Adrian Bunk75c96f82005-05-05 16:16:09 -07003071static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3072 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073{
Sage Weil51892bb2011-05-24 13:06:13 -07003074 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 int error;
3076
3077 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3078 if (error)
3079 return error;
3080
3081 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003083 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003084
3085 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003087 goto out;
3088
3089 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3090 if (error)
3091 goto out;
3092
3093 if (target)
3094 dont_mount(new_dentry);
3095 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3096 d_move(old_dentry, new_dentry);
3097out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003099 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 dput(new_dentry);
3101 return error;
3102}
3103
3104int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3105 struct inode *new_dir, struct dentry *new_dentry)
3106{
3107 int error;
3108 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003109 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
3111 if (old_dentry->d_inode == new_dentry->d_inode)
3112 return 0;
3113
3114 error = may_delete(old_dir, old_dentry, is_dir);
3115 if (error)
3116 return error;
3117
3118 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003119 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 else
3121 error = may_delete(new_dir, new_dentry, is_dir);
3122 if (error)
3123 return error;
3124
Al Viroacfa4382008-12-04 10:06:33 -05003125 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 return -EPERM;
3127
Robert Love0eeca282005-07-12 17:06:03 -04003128 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3129
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 if (is_dir)
3131 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3132 else
3133 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003134 if (!error)
3135 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003136 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003137 fsnotify_oldname_free(old_name);
3138
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 return error;
3140}
3141
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003142SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3143 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144{
Al Viro2ad94ae2008-07-21 09:32:51 -04003145 struct dentry *old_dir, *new_dir;
3146 struct dentry *old_dentry, *new_dentry;
3147 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003149 char *from;
3150 char *to;
3151 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152
Al Viro2ad94ae2008-07-21 09:32:51 -04003153 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 if (error)
3155 goto exit;
3156
Al Viro2ad94ae2008-07-21 09:32:51 -04003157 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 if (error)
3159 goto exit1;
3160
3161 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003162 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 goto exit2;
3164
Jan Blunck4ac91372008-02-14 19:34:32 -08003165 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 error = -EBUSY;
3167 if (oldnd.last_type != LAST_NORM)
3168 goto exit2;
3169
Jan Blunck4ac91372008-02-14 19:34:32 -08003170 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 if (newnd.last_type != LAST_NORM)
3172 goto exit2;
3173
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003174 oldnd.flags &= ~LOOKUP_PARENT;
3175 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003176 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003177
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 trap = lock_rename(new_dir, old_dir);
3179
Christoph Hellwig49705b72005-11-08 21:35:06 -08003180 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 error = PTR_ERR(old_dentry);
3182 if (IS_ERR(old_dentry))
3183 goto exit3;
3184 /* source must exist */
3185 error = -ENOENT;
3186 if (!old_dentry->d_inode)
3187 goto exit4;
3188 /* unless the source is a directory trailing slashes give -ENOTDIR */
3189 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3190 error = -ENOTDIR;
3191 if (oldnd.last.name[oldnd.last.len])
3192 goto exit4;
3193 if (newnd.last.name[newnd.last.len])
3194 goto exit4;
3195 }
3196 /* source should not be ancestor of target */
3197 error = -EINVAL;
3198 if (old_dentry == trap)
3199 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003200 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201 error = PTR_ERR(new_dentry);
3202 if (IS_ERR(new_dentry))
3203 goto exit4;
3204 /* target should not be an ancestor of source */
3205 error = -ENOTEMPTY;
3206 if (new_dentry == trap)
3207 goto exit5;
3208
Dave Hansen9079b1e2008-02-15 14:37:49 -08003209 error = mnt_want_write(oldnd.path.mnt);
3210 if (error)
3211 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003212 error = security_path_rename(&oldnd.path, old_dentry,
3213 &newnd.path, new_dentry);
3214 if (error)
3215 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 error = vfs_rename(old_dir->d_inode, old_dentry,
3217 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003218exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003219 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220exit5:
3221 dput(new_dentry);
3222exit4:
3223 dput(old_dentry);
3224exit3:
3225 unlock_rename(new_dir, old_dir);
3226exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003227 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003228 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003230 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003232exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 return error;
3234}
3235
Heiko Carstensa26eab22009-01-14 14:14:17 +01003236SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003237{
3238 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3239}
3240
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3242{
3243 int len;
3244
3245 len = PTR_ERR(link);
3246 if (IS_ERR(link))
3247 goto out;
3248
3249 len = strlen(link);
3250 if (len > (unsigned) buflen)
3251 len = buflen;
3252 if (copy_to_user(buffer, link, len))
3253 len = -EFAULT;
3254out:
3255 return len;
3256}
3257
3258/*
3259 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3260 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3261 * using) it for any given inode is up to filesystem.
3262 */
3263int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3264{
3265 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003266 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003267 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003268
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003270 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003271 if (IS_ERR(cookie))
3272 return PTR_ERR(cookie);
3273
3274 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3275 if (dentry->d_inode->i_op->put_link)
3276 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3277 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278}
3279
3280int vfs_follow_link(struct nameidata *nd, const char *link)
3281{
3282 return __vfs_follow_link(nd, link);
3283}
3284
3285/* get the link contents into pagecache */
3286static char *page_getlink(struct dentry * dentry, struct page **ppage)
3287{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003288 char *kaddr;
3289 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003291 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003293 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003295 kaddr = kmap(page);
3296 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3297 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298}
3299
3300int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3301{
3302 struct page *page = NULL;
3303 char *s = page_getlink(dentry, &page);
3304 int res = vfs_readlink(dentry,buffer,buflen,s);
3305 if (page) {
3306 kunmap(page);
3307 page_cache_release(page);
3308 }
3309 return res;
3310}
3311
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003312void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003314 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003316 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317}
3318
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003319void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003321 struct page *page = cookie;
3322
3323 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 kunmap(page);
3325 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 }
3327}
3328
Nick Piggin54566b22009-01-04 12:00:53 -08003329/*
3330 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3331 */
3332int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333{
3334 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003335 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003336 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003337 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003339 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3340 if (nofs)
3341 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
NeilBrown7e53cac2006-03-25 03:07:57 -08003343retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003344 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003345 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003347 goto fail;
3348
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 kaddr = kmap_atomic(page, KM_USER0);
3350 memcpy(kaddr, symname, len-1);
3351 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003352
3353 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3354 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 if (err < 0)
3356 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003357 if (err < len-1)
3358 goto retry;
3359
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 mark_inode_dirty(inode);
3361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362fail:
3363 return err;
3364}
3365
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003366int page_symlink(struct inode *inode, const char *symname, int len)
3367{
3368 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003369 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003370}
3371
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003372const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 .readlink = generic_readlink,
3374 .follow_link = page_follow_link_light,
3375 .put_link = page_put_link,
3376};
3377
Al Viro2d8f3032008-07-22 09:59:21 -04003378EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003379EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380EXPORT_SYMBOL(follow_down);
3381EXPORT_SYMBOL(follow_up);
3382EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3383EXPORT_SYMBOL(getname);
3384EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385EXPORT_SYMBOL(lookup_one_len);
3386EXPORT_SYMBOL(page_follow_link_light);
3387EXPORT_SYMBOL(page_put_link);
3388EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003389EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390EXPORT_SYMBOL(page_symlink);
3391EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003392EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003393EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003394EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395EXPORT_SYMBOL(unlock_rename);
3396EXPORT_SYMBOL(vfs_create);
3397EXPORT_SYMBOL(vfs_follow_link);
3398EXPORT_SYMBOL(vfs_link);
3399EXPORT_SYMBOL(vfs_mkdir);
3400EXPORT_SYMBOL(vfs_mknod);
3401EXPORT_SYMBOL(generic_permission);
3402EXPORT_SYMBOL(vfs_readlink);
3403EXPORT_SYMBOL(vfs_rename);
3404EXPORT_SYMBOL(vfs_rmdir);
3405EXPORT_SYMBOL(vfs_symlink);
3406EXPORT_SYMBOL(vfs_unlink);
3407EXPORT_SYMBOL(dentry_unhash);
3408EXPORT_SYMBOL(generic_readlink);