blob: 044215a7bb0c20bbe61e9ae7062fde3b1ec7f592 [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>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070036#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <asm/uaccess.h>
38
Eric Parise81e3f42009-12-04 15:47:36 -050039#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050040#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
47 *
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
54 *
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
58 *
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
61 *
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
68 */
69
70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030076 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 *
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
85 */
86
87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
89 *
90 * [10-Sep-98 Alan Modra] Another symlink change.
91 */
92
93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
100 *
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
106 */
107/*
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * any extra contention...
111 */
112
113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100120static char *getname_flags(const char __user *filename, int flags, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700122 char *result = __getname(), *err;
123 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700125 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500126 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700128 len = strncpy_from_user(result, filename, PATH_MAX);
129 err = ERR_PTR(len);
130 if (unlikely(len < 0))
131 goto error;
132
133 /* The empty path is special. */
134 if (unlikely(!len)) {
135 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500136 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
139 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700141
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
145 return result;
146 }
147
148error:
149 __putname(result);
150 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Al Virof52e0c12011-03-14 18:56:51 -0400153char *getname(const char __user * filename)
154{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700155 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400161 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
Linus Torvaldse77819e2011-07-22 19:30:19 -0700169static int check_acl(struct inode *inode, int mask)
170{
Linus Torvalds84635d62011-07-25 22:47:03 -0700171#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700172 struct posix_acl *acl;
173
Linus Torvaldse77819e2011-07-22 19:30:19 -0700174 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
176 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700177 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
180 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700182 }
183
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
185
186 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700190 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700193 */
194 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
197 if (IS_ERR(acl))
198 return PTR_ERR(acl);
199 } else {
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
201 return -EAGAIN;
202 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700203 }
204
205 if (acl) {
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
208 return error;
209 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700210#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700211
212 return -EAGAIN;
213}
214
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700215/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530216 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700217 */
Al Viro7e401452011-06-20 19:12:17 -0400218static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700219{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700220 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700221
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700223 mode >>= 6;
224 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400226 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100227 if (error != -EAGAIN)
228 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229 }
230
231 if (in_group_p(inode->i_gid))
232 mode >>= 3;
233 }
234
235 /*
236 * If the DACs are ok we don't need any capability check.
237 */
Al Viro9c2c7032011-06-20 19:06:22 -0400238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700239 return 0;
240 return -EACCES;
241}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100244 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 *
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100251 * are used for other things.
252 *
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 */
Al Viro2830ba72011-06-20 19:16:29 -0400257int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700259 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530262 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
Al Viro7e401452011-06-20 19:12:17 -0400264 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700265 if (ret != -EACCES)
266 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Al Virod594e7e2011-06-20 19:55:42 -0400268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400271 return 0;
272 if (!(mask & MAY_WRITE))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400274 return 0;
275 return -EACCES;
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 /*
278 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 */
Al Virod594e7e2011-06-20 19:55:42 -0400282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return 0;
285
286 /*
287 * Searching includes executable on directories, else just read.
288 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400290 if (mask == MAY_READ)
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 return 0;
293
294 return -EACCES;
295}
296
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700297/*
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
302 */
303static inline int do_inode_permission(struct inode *inode, int mask)
304{
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
308
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
313 }
314 return generic_permission(inode, mask);
315}
316
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200317/**
318 * inode_permission - check for access rights to a given inode
319 * @inode: inode to check permission on
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200321 *
322 * Used to check for read/write/execute permissions on an inode.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530326 *
327 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200328 */
Al Virof419a2e2008-07-22 00:07:17 -0400329int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Al Viroe6305c42008-07-15 21:03:57 -0400331 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700333 if (unlikely(mask & MAY_WRITE)) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700334 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 /*
337 * Nobody gets write access to a read-only fs.
338 */
339 if (IS_RDONLY(inode) &&
340 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
341 return -EROFS;
342
343 /*
344 * Nobody gets write access to an immutable file.
345 */
346 if (IS_IMMUTABLE(inode))
347 return -EACCES;
348 }
349
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700350 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (retval)
352 return retval;
353
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700354 retval = devcgroup_inode_permission(inode, mask);
355 if (retval)
356 return retval;
357
Eric Parisd09ca732010-07-23 11:43:57 -0400358 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
360
Al Virof4d6ff82011-06-19 13:14:21 -0400361/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800362 * path_get - get a reference to a path
363 * @path: path to get the reference to
364 *
365 * Given a path increment the reference count to the dentry and the vfsmount.
366 */
367void path_get(struct path *path)
368{
369 mntget(path->mnt);
370 dget(path->dentry);
371}
372EXPORT_SYMBOL(path_get);
373
374/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800375 * path_put - put a reference to a path
376 * @path: path to put the reference to
377 *
378 * Given a path decrement the reference count to the dentry and the vfsmount.
379 */
380void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Jan Blunck1d957f92008-02-14 19:34:35 -0800382 dput(path->dentry);
383 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
Jan Blunck1d957f92008-02-14 19:34:35 -0800385EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Al Viro19660af2011-03-25 10:32:48 -0400387/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100388 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400389 * Documentation/filesystems/path-lookup.txt). In situations when we can't
390 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
391 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
392 * mode. Refcounts are grabbed at the last known good point before rcu-walk
393 * got stuck, so ref-walk may continue from there. If this is not successful
394 * (eg. a seqcount has changed), then failure is returned and it's up to caller
395 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100396 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100397
398/**
Al Viro19660af2011-03-25 10:32:48 -0400399 * unlazy_walk - try to switch to ref-walk mode.
400 * @nd: nameidata pathwalk data
401 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800402 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100403 *
Al Viro19660af2011-03-25 10:32:48 -0400404 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
405 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
406 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100407 */
Al Viro19660af2011-03-25 10:32:48 -0400408static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100409{
410 struct fs_struct *fs = current->fs;
411 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500412 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100413
414 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500415 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
416 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100417 spin_lock(&fs->lock);
418 if (nd->root.mnt != fs->root.mnt ||
419 nd->root.dentry != fs->root.dentry)
420 goto err_root;
421 }
422 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400423 if (!dentry) {
424 if (!__d_rcu_to_refcount(parent, nd->seq))
425 goto err_parent;
426 BUG_ON(nd->inode != parent->d_inode);
427 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400428 if (dentry->d_parent != parent)
429 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400430 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
431 if (!__d_rcu_to_refcount(dentry, nd->seq))
432 goto err_child;
433 /*
434 * If the sequence check on the child dentry passed, then
435 * the child has not been removed from its parent. This
436 * means the parent dentry must be valid and able to take
437 * a reference at this point.
438 */
439 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
440 BUG_ON(!parent->d_count);
441 parent->d_count++;
442 spin_unlock(&dentry->d_lock);
443 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100444 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500445 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100446 path_get(&nd->root);
447 spin_unlock(&fs->lock);
448 }
449 mntget(nd->path.mnt);
450
451 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930452 br_read_unlock(&vfsmount_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100453 nd->flags &= ~LOOKUP_RCU;
454 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400455
456err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100457 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400458err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100459 spin_unlock(&parent->d_lock);
460err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500461 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100462 spin_unlock(&fs->lock);
463 return -ECHILD;
464}
465
Al Virof60aef72011-02-15 01:35:28 -0500466static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100467{
Al Virof60aef72011-02-15 01:35:28 -0500468 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100469}
470
Al Viro9f1fafe2011-03-25 11:00:12 -0400471/**
472 * complete_walk - successful completion of path walk
473 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500474 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400475 * If we had been in RCU mode, drop out of it and legitimize nd->path.
476 * Revalidate the final result, unless we'd already done that during
477 * the path walk or the filesystem doesn't ask for it. Return 0 on
478 * success, -error on failure. In case of failure caller does not
479 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500480 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400481static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500482{
Al Viro16c2cd72011-02-22 15:50:10 -0500483 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500484 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500485
Al Viro9f1fafe2011-03-25 11:00:12 -0400486 if (nd->flags & LOOKUP_RCU) {
487 nd->flags &= ~LOOKUP_RCU;
488 if (!(nd->flags & LOOKUP_ROOT))
489 nd->root.mnt = NULL;
490 spin_lock(&dentry->d_lock);
491 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
492 spin_unlock(&dentry->d_lock);
493 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930494 br_read_unlock(&vfsmount_lock);
Al Viro9f1fafe2011-03-25 11:00:12 -0400495 return -ECHILD;
496 }
497 BUG_ON(nd->inode != dentry->d_inode);
498 spin_unlock(&dentry->d_lock);
499 mntget(nd->path.mnt);
500 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930501 br_read_unlock(&vfsmount_lock);
Al Viro9f1fafe2011-03-25 11:00:12 -0400502 }
503
Al Viro16c2cd72011-02-22 15:50:10 -0500504 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500505 return 0;
506
Al Viro16c2cd72011-02-22 15:50:10 -0500507 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
508 return 0;
509
510 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
511 return 0;
512
513 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100514 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500515 if (status > 0)
516 return 0;
517
Al Viro16c2cd72011-02-22 15:50:10 -0500518 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500519 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500520
Al Viro9f1fafe2011-03-25 11:00:12 -0400521 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500522 return status;
523}
524
Al Viro2a737872009-04-07 11:49:53 -0400525static __always_inline void set_root(struct nameidata *nd)
526{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200527 if (!nd->root.mnt)
528 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400529}
530
Al Viro6de88d72009-08-09 01:41:57 +0400531static int link_path_walk(const char *, struct nameidata *);
532
Nick Piggin31e6b012011-01-07 17:49:52 +1100533static __always_inline void set_root_rcu(struct nameidata *nd)
534{
535 if (!nd->root.mnt) {
536 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100537 unsigned seq;
538
539 do {
540 seq = read_seqcount_begin(&fs->seq);
541 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700542 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100543 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100544 }
545}
546
Arjan van de Venf1662352006-01-14 13:21:31 -0800547static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Nick Piggin31e6b012011-01-07 17:49:52 +1100549 int ret;
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (IS_ERR(link))
552 goto fail;
553
554 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400555 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800556 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400557 nd->path = nd->root;
558 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500559 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100561 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100562
Nick Piggin31e6b012011-01-07 17:49:52 +1100563 ret = link_path_walk(link, nd);
564 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800566 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 return PTR_ERR(link);
568}
569
Jan Blunck1d957f92008-02-14 19:34:35 -0800570static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700571{
572 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800573 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700574 mntput(path->mnt);
575}
576
Nick Piggin7b9337a2011-01-14 08:42:43 +0000577static inline void path_to_nameidata(const struct path *path,
578 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700579{
Nick Piggin31e6b012011-01-07 17:49:52 +1100580 if (!(nd->flags & LOOKUP_RCU)) {
581 dput(nd->path.dentry);
582 if (nd->path.mnt != path->mnt)
583 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800584 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100585 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800586 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700587}
588
Al Viro574197e2011-03-14 22:20:34 -0400589static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
590{
591 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400592 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400593 inode->i_op->put_link(link->dentry, nd, cookie);
594 path_put(link);
595}
596
Al Virodef4af32009-12-26 08:37:05 -0500597static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400598follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800599{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000600 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400601 int error;
602 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800603
Al Viro844a3912011-02-15 00:38:26 -0500604 BUG_ON(nd->flags & LOOKUP_RCU);
605
Al Viro0e794582011-03-16 02:45:02 -0400606 if (link->mnt == nd->path.mnt)
607 mntget(link->mnt);
608
Al Viro6d7b5aa2012-06-10 04:15:17 -0400609 error = -ELOOP;
610 if (unlikely(current->total_link_count >= 40))
611 goto out_put_nd_path;
612
Al Viro574197e2011-03-14 22:20:34 -0400613 cond_resched();
614 current->total_link_count++;
615
Al Viro68ac1232012-03-15 08:21:57 -0400616 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800617 nd_set_link(nd, NULL);
618
Al Viro36f3b4f2011-02-22 21:24:38 -0500619 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400620 if (error)
621 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500622
Al Viro86acdca12009-12-22 23:45:11 -0500623 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500624 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
625 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400626 if (IS_ERR(*p))
627 goto out_put_link;
628
629 error = 0;
630 s = nd_get_link(nd);
631 if (s) {
632 error = __vfs_follow_link(nd, s);
633 } else if (nd->last_type == LAST_BIND) {
634 nd->flags |= LOOKUP_JUMPED;
635 nd->inode = nd->path.dentry->d_inode;
636 if (nd->inode->i_op->follow_link) {
637 /* stepped on a _really_ weird one */
638 path_put(&nd->path);
639 error = -ELOOP;
Al Virobcda7652011-03-13 16:42:14 -0400640 }
Ian Kent051d3812006-03-27 01:14:53 -0800641 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400642 if (unlikely(error))
643 put_link(nd, link, *p);
644
645 return error;
646
647out_put_nd_path:
648 path_put(&nd->path);
649out_put_link:
650 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800651 return error;
652}
653
Nick Piggin31e6b012011-01-07 17:49:52 +1100654static int follow_up_rcu(struct path *path)
655{
Al Viro0714a532011-11-24 22:19:58 -0500656 struct mount *mnt = real_mount(path->mnt);
657 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100658 struct dentry *mountpoint;
659
Al Viro0714a532011-11-24 22:19:58 -0500660 parent = mnt->mnt_parent;
661 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100662 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500663 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100664 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500665 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100666 return 1;
667}
668
Al Virobab77eb2009-04-18 03:26:48 -0400669int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
Al Viro0714a532011-11-24 22:19:58 -0500671 struct mount *mnt = real_mount(path->mnt);
672 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000674
Andi Kleen962830d2012-05-08 13:32:02 +0930675 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500676 parent = mnt->mnt_parent;
677 if (&parent->mnt == path->mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930678 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return 0;
680 }
Al Viro0714a532011-11-24 22:19:58 -0500681 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500682 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930683 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400684 dput(path->dentry);
685 path->dentry = mountpoint;
686 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500687 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return 1;
689}
690
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100691/*
David Howells9875cf82011-01-14 18:45:21 +0000692 * Perform an automount
693 * - return -EISDIR to tell follow_managed() to stop and return the path we
694 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 */
David Howells9875cf82011-01-14 18:45:21 +0000696static int follow_automount(struct path *path, unsigned flags,
697 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100698{
David Howells9875cf82011-01-14 18:45:21 +0000699 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000700 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100701
David Howells9875cf82011-01-14 18:45:21 +0000702 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
703 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700704
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200705 /* We don't want to mount if someone's just doing a stat -
706 * unless they're stat'ing a directory and appended a '/' to
707 * the name.
708 *
709 * We do, however, want to mount if someone wants to open or
710 * create a file of any type under the mountpoint, wants to
711 * traverse through the mountpoint or wants to open the
712 * mounted directory. Also, autofs may mark negative dentries
713 * as being automount points. These will need the attentions
714 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000715 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200716 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700717 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200718 path->dentry->d_inode)
719 return -EISDIR;
720
David Howells9875cf82011-01-14 18:45:21 +0000721 current->total_link_count++;
722 if (current->total_link_count >= 40)
723 return -ELOOP;
724
725 mnt = path->dentry->d_op->d_automount(path);
726 if (IS_ERR(mnt)) {
727 /*
728 * The filesystem is allowed to return -EISDIR here to indicate
729 * it doesn't want to automount. For instance, autofs would do
730 * this so that its userspace daemon can mount on this dentry.
731 *
732 * However, we can only permit this if it's a terminal point in
733 * the path being looked up; if it wasn't then the remainder of
734 * the path is inaccessible and we should say so.
735 */
Al Viro49084c32011-06-25 21:59:52 -0400736 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000737 return -EREMOTE;
738 return PTR_ERR(mnt);
739 }
David Howellsea5b7782011-01-14 19:10:03 +0000740
David Howells9875cf82011-01-14 18:45:21 +0000741 if (!mnt) /* mount collision */
742 return 0;
743
Al Viro8aef1882011-06-16 15:10:06 +0100744 if (!*need_mntput) {
745 /* lock_mount() may release path->mnt on error */
746 mntget(path->mnt);
747 *need_mntput = true;
748 }
Al Viro19a167a2011-01-17 01:35:23 -0500749 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000750
David Howellsea5b7782011-01-14 19:10:03 +0000751 switch (err) {
752 case -EBUSY:
753 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500754 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000755 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100756 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000757 path->mnt = mnt;
758 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000759 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500760 default:
761 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000762 }
Al Viro19a167a2011-01-17 01:35:23 -0500763
David Howells9875cf82011-01-14 18:45:21 +0000764}
765
766/*
767 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000768 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000769 * - Flagged as mountpoint
770 * - Flagged as automount point
771 *
772 * This may only be called in refwalk mode.
773 *
774 * Serialization is taken care of in namespace.c
775 */
776static int follow_managed(struct path *path, unsigned flags)
777{
Al Viro8aef1882011-06-16 15:10:06 +0100778 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000779 unsigned managed;
780 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100781 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000782
783 /* Given that we're not holding a lock here, we retain the value in a
784 * local variable for each dentry as we look at it so that we don't see
785 * the components of that value change under us */
786 while (managed = ACCESS_ONCE(path->dentry->d_flags),
787 managed &= DCACHE_MANAGED_DENTRY,
788 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000789 /* Allow the filesystem to manage the transit without i_mutex
790 * being held. */
791 if (managed & DCACHE_MANAGE_TRANSIT) {
792 BUG_ON(!path->dentry->d_op);
793 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400794 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000795 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100796 break;
David Howellscc53ce52011-01-14 18:45:26 +0000797 }
798
David Howells9875cf82011-01-14 18:45:21 +0000799 /* Transit to a mounted filesystem. */
800 if (managed & DCACHE_MOUNTED) {
801 struct vfsmount *mounted = lookup_mnt(path);
802 if (mounted) {
803 dput(path->dentry);
804 if (need_mntput)
805 mntput(path->mnt);
806 path->mnt = mounted;
807 path->dentry = dget(mounted->mnt_root);
808 need_mntput = true;
809 continue;
810 }
811
812 /* Something is mounted on this dentry in another
813 * namespace and/or whatever was mounted there in this
814 * namespace got unmounted before we managed to get the
815 * vfsmount_lock */
816 }
817
818 /* Handle an automount point */
819 if (managed & DCACHE_NEED_AUTOMOUNT) {
820 ret = follow_automount(path, flags, &need_mntput);
821 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100822 break;
David Howells9875cf82011-01-14 18:45:21 +0000823 continue;
824 }
825
826 /* We didn't change the current path point */
827 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 }
Al Viro8aef1882011-06-16 15:10:06 +0100829
830 if (need_mntput && path->mnt == mnt)
831 mntput(path->mnt);
832 if (ret == -EISDIR)
833 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +0000834 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
836
David Howellscc53ce52011-01-14 18:45:26 +0000837int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
839 struct vfsmount *mounted;
840
Al Viro1c755af2009-04-18 14:06:57 -0400841 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400843 dput(path->dentry);
844 mntput(path->mnt);
845 path->mnt = mounted;
846 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return 1;
848 }
849 return 0;
850}
851
Ian Kent62a73752011-03-25 01:51:02 +0800852static inline bool managed_dentry_might_block(struct dentry *dentry)
853{
854 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
855 dentry->d_op->d_manage(dentry, true) < 0);
856}
857
David Howells9875cf82011-01-14 18:45:21 +0000858/*
Al Viro287548e2011-05-27 06:50:06 -0400859 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
860 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000861 */
862static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400863 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000864{
Ian Kent62a73752011-03-25 01:51:02 +0800865 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -0500866 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800867 /*
868 * Don't forget we might have a non-mountpoint managed dentry
869 * that wants to block transit.
870 */
Al Viro287548e2011-05-27 06:50:06 -0400871 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000872 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800873
874 if (!d_mountpoint(path->dentry))
875 break;
876
David Howells9875cf82011-01-14 18:45:21 +0000877 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
878 if (!mounted)
879 break;
Al Viroc7105362011-11-24 18:22:03 -0500880 path->mnt = &mounted->mnt;
881 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +0000882 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +0000883 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700884 /*
885 * Update the inode too. We don't need to re-check the
886 * dentry sequence number here after this d_inode read,
887 * because a mount-point is always pinned.
888 */
889 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000890 }
David Howells9875cf82011-01-14 18:45:21 +0000891 return true;
892}
893
Al Virodea39372011-05-27 06:53:39 -0400894static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400895{
Al Virodea39372011-05-27 06:53:39 -0400896 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -0500897 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400898 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400899 if (!mounted)
900 break;
Al Viroc7105362011-11-24 18:22:03 -0500901 nd->path.mnt = &mounted->mnt;
902 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -0400903 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400904 }
905}
906
Nick Piggin31e6b012011-01-07 17:49:52 +1100907static int follow_dotdot_rcu(struct nameidata *nd)
908{
Nick Piggin31e6b012011-01-07 17:49:52 +1100909 set_root_rcu(nd);
910
David Howells9875cf82011-01-14 18:45:21 +0000911 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100912 if (nd->path.dentry == nd->root.dentry &&
913 nd->path.mnt == nd->root.mnt) {
914 break;
915 }
916 if (nd->path.dentry != nd->path.mnt->mnt_root) {
917 struct dentry *old = nd->path.dentry;
918 struct dentry *parent = old->d_parent;
919 unsigned seq;
920
921 seq = read_seqcount_begin(&parent->d_seq);
922 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500923 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100924 nd->path.dentry = parent;
925 nd->seq = seq;
926 break;
927 }
928 if (!follow_up_rcu(&nd->path))
929 break;
930 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100931 }
Al Virodea39372011-05-27 06:53:39 -0400932 follow_mount_rcu(nd);
933 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100934 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500935
936failed:
937 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500938 if (!(nd->flags & LOOKUP_ROOT))
939 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500940 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930941 br_read_unlock(&vfsmount_lock);
Al Viroef7562d2011-03-04 14:35:59 -0500942 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100943}
944
David Howells9875cf82011-01-14 18:45:21 +0000945/*
David Howellscc53ce52011-01-14 18:45:26 +0000946 * Follow down to the covering mount currently visible to userspace. At each
947 * point, the filesystem owning that dentry may be queried as to whether the
948 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000949 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400950int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000951{
952 unsigned managed;
953 int ret;
954
955 while (managed = ACCESS_ONCE(path->dentry->d_flags),
956 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
957 /* Allow the filesystem to manage the transit without i_mutex
958 * being held.
959 *
960 * We indicate to the filesystem if someone is trying to mount
961 * something here. This gives autofs the chance to deny anyone
962 * other than its daemon the right to mount on its
963 * superstructure.
964 *
965 * The filesystem may sleep at this point.
966 */
967 if (managed & DCACHE_MANAGE_TRANSIT) {
968 BUG_ON(!path->dentry->d_op);
969 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000970 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400971 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000972 if (ret < 0)
973 return ret == -EISDIR ? 0 : ret;
974 }
975
976 /* Transit to a mounted filesystem. */
977 if (managed & DCACHE_MOUNTED) {
978 struct vfsmount *mounted = lookup_mnt(path);
979 if (!mounted)
980 break;
981 dput(path->dentry);
982 mntput(path->mnt);
983 path->mnt = mounted;
984 path->dentry = dget(mounted->mnt_root);
985 continue;
986 }
987
988 /* Don't handle automount points here */
989 break;
990 }
991 return 0;
992}
993
994/*
David Howells9875cf82011-01-14 18:45:21 +0000995 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
996 */
997static void follow_mount(struct path *path)
998{
999 while (d_mountpoint(path->dentry)) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1001 if (!mounted)
1002 break;
1003 dput(path->dentry);
1004 mntput(path->mnt);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1007 }
1008}
1009
Nick Piggin31e6b012011-01-07 17:49:52 +11001010static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
Al Viro2a737872009-04-07 11:49:53 -04001012 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001015 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Al Viro2a737872009-04-07 11:49:53 -04001017 if (nd->path.dentry == nd->root.dentry &&
1018 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 break;
1020 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001021 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001022 /* rare case of legitimate dget_parent()... */
1023 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 dput(old);
1025 break;
1026 }
Al Viro3088dd72010-01-30 15:47:29 -05001027 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
Al Viro79ed0222009-04-18 13:59:41 -04001030 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001031 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032}
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001035 * This looks up the name in dcache, possibly revalidates the old dentry and
1036 * allocates a new one if not found or not valid. In the need_lookup argument
1037 * returns whether i_op->lookup is necessary.
1038 *
1039 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001040 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001041static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1042 struct nameidata *nd, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001043{
Nick Pigginbaa03892010-08-18 04:37:31 +10001044 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001045 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001046
Miklos Szeredibad61182012-03-26 12:54:24 +02001047 *need_lookup = false;
1048 dentry = d_lookup(dir, name);
1049 if (dentry) {
1050 if (d_need_lookup(dentry)) {
1051 *need_lookup = true;
1052 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1053 error = d_revalidate(dentry, nd);
1054 if (unlikely(error <= 0)) {
1055 if (error < 0) {
1056 dput(dentry);
1057 return ERR_PTR(error);
1058 } else if (!d_invalidate(dentry)) {
1059 dput(dentry);
1060 dentry = NULL;
1061 }
1062 }
1063 }
1064 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001065
Miklos Szeredibad61182012-03-26 12:54:24 +02001066 if (!dentry) {
1067 dentry = d_alloc(dir, name);
1068 if (unlikely(!dentry))
1069 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001070
Miklos Szeredibad61182012-03-26 12:54:24 +02001071 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001072 }
1073 return dentry;
1074}
1075
1076/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001077 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1078 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1079 *
1080 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001081 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001082static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1083 struct nameidata *nd)
Josef Bacik44396f42011-05-31 11:58:49 -04001084{
Josef Bacik44396f42011-05-31 11:58:49 -04001085 struct dentry *old;
1086
1087 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001088 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001089 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001090 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001091 }
Josef Bacik44396f42011-05-31 11:58:49 -04001092
Miklos Szeredibad61182012-03-26 12:54:24 +02001093 old = dir->i_op->lookup(dir, dentry, nd);
Josef Bacik44396f42011-05-31 11:58:49 -04001094 if (unlikely(old)) {
1095 dput(dentry);
1096 dentry = old;
1097 }
1098 return dentry;
1099}
1100
Al Viroa3255542012-03-30 14:41:51 -04001101static struct dentry *__lookup_hash(struct qstr *name,
1102 struct dentry *base, struct nameidata *nd)
1103{
Miklos Szeredibad61182012-03-26 12:54:24 +02001104 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001105 struct dentry *dentry;
1106
Miklos Szeredibad61182012-03-26 12:54:24 +02001107 dentry = lookup_dcache(name, base, nd, &need_lookup);
1108 if (!need_lookup)
1109 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001110
Miklos Szeredibad61182012-03-26 12:54:24 +02001111 return lookup_real(base->d_inode, dentry, nd);
Al Viroa3255542012-03-30 14:41:51 -04001112}
1113
Josef Bacik44396f42011-05-31 11:58:49 -04001114/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 * It's more convoluted than I'd like it to be, but... it's still fairly
1116 * small and for now I'd prefer to have fast path as straight as possible.
1117 * It _is_ time-critical.
1118 */
Miklos Szeredi697f5142012-05-21 17:30:05 +02001119static int lookup_fast(struct nameidata *nd, struct qstr *name,
1120 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
Jan Blunck4ac91372008-02-14 19:34:32 -08001122 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001123 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001124 int need_reval = 1;
1125 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001126 int err;
1127
Al Viro3cac2602009-08-13 18:27:43 +04001128 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001129 * Rename seqlock is not required here because in the off chance
1130 * of a false negative due to a concurrent rename, we're going to
1131 * do the non-racy lookup, below.
1132 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001133 if (nd->flags & LOOKUP_RCU) {
1134 unsigned seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001135 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001136 if (!dentry)
1137 goto unlazy;
1138
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001139 /*
1140 * This sequence count validates that the inode matches
1141 * the dentry name information from lookup.
1142 */
1143 *inode = dentry->d_inode;
1144 if (read_seqcount_retry(&dentry->d_seq, seq))
1145 return -ECHILD;
1146
1147 /*
1148 * This sequence count validates that the parent had no
1149 * changes while we did the lookup of the dentry above.
1150 *
1151 * The memory barrier in read_seqcount_begin of child is
1152 * enough, we can use __read_seqcount_retry here.
1153 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001154 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1155 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001156 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001157
Miklos Szeredifa4ee152012-03-26 12:54:19 +02001158 if (unlikely(d_need_lookup(dentry)))
1159 goto unlazy;
Al Viro24643082011-02-15 01:26:22 -05001160 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001161 status = d_revalidate(dentry, nd);
1162 if (unlikely(status <= 0)) {
1163 if (status != -ECHILD)
1164 need_reval = 0;
1165 goto unlazy;
1166 }
Al Viro24643082011-02-15 01:26:22 -05001167 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001168 path->mnt = mnt;
1169 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001170 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1171 goto unlazy;
1172 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1173 goto unlazy;
1174 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001175unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001176 if (unlazy_walk(nd, dentry))
1177 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001178 } else {
1179 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001180 }
Al Viro5a18fff2011-03-11 04:44:53 -05001181
Al Viro81e6f522012-03-30 14:48:04 -04001182 if (unlikely(!dentry))
1183 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001184
Al Viro81e6f522012-03-30 14:48:04 -04001185 if (unlikely(d_need_lookup(dentry))) {
1186 dput(dentry);
1187 goto need_lookup;
Al Viro24643082011-02-15 01:26:22 -05001188 }
Al Viro81e6f522012-03-30 14:48:04 -04001189
Al Viro5a18fff2011-03-11 04:44:53 -05001190 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1191 status = d_revalidate(dentry, nd);
1192 if (unlikely(status <= 0)) {
1193 if (status < 0) {
1194 dput(dentry);
1195 return status;
1196 }
1197 if (!d_invalidate(dentry)) {
1198 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001199 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001200 }
1201 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001202
David Howells9875cf82011-01-14 18:45:21 +00001203 path->mnt = mnt;
1204 path->dentry = dentry;
1205 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001206 if (unlikely(err < 0)) {
1207 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001208 return err;
Ian Kent89312212011-01-18 12:06:10 +08001209 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001210 if (err)
1211 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001212 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001214
1215need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001216 return 1;
1217}
1218
1219/* Fast lookup failed, do it the slow way */
1220static int lookup_slow(struct nameidata *nd, struct qstr *name,
1221 struct path *path)
1222{
1223 struct dentry *dentry, *parent;
1224 int err;
1225
1226 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001227 BUG_ON(nd->inode != parent->d_inode);
1228
1229 mutex_lock(&parent->d_inode->i_mutex);
1230 dentry = __lookup_hash(name, parent, nd);
1231 mutex_unlock(&parent->d_inode->i_mutex);
1232 if (IS_ERR(dentry))
1233 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001234 path->mnt = nd->path.mnt;
1235 path->dentry = dentry;
1236 err = follow_managed(path, nd->flags);
1237 if (unlikely(err < 0)) {
1238 path_put_conditional(path, nd);
1239 return err;
1240 }
1241 if (err)
1242 nd->flags |= LOOKUP_JUMPED;
1243 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244}
1245
Al Viro52094c82011-02-21 21:34:47 -05001246static inline int may_lookup(struct nameidata *nd)
1247{
1248 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001249 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001250 if (err != -ECHILD)
1251 return err;
Al Viro19660af2011-03-25 10:32:48 -04001252 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001253 return -ECHILD;
1254 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001255 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001256}
1257
Al Viro9856fa12011-03-04 14:22:06 -05001258static inline int handle_dots(struct nameidata *nd, int type)
1259{
1260 if (type == LAST_DOTDOT) {
1261 if (nd->flags & LOOKUP_RCU) {
1262 if (follow_dotdot_rcu(nd))
1263 return -ECHILD;
1264 } else
1265 follow_dotdot(nd);
1266 }
1267 return 0;
1268}
1269
Al Viro951361f2011-03-04 14:44:37 -05001270static void terminate_walk(struct nameidata *nd)
1271{
1272 if (!(nd->flags & LOOKUP_RCU)) {
1273 path_put(&nd->path);
1274 } else {
1275 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001276 if (!(nd->flags & LOOKUP_ROOT))
1277 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001278 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301279 br_read_unlock(&vfsmount_lock);
Al Viro951361f2011-03-04 14:44:37 -05001280 }
1281}
1282
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001283/*
1284 * Do we need to follow links? We _really_ want to be able
1285 * to do this check without having to look at inode->i_op,
1286 * so we keep a cache of "no, this doesn't need follow_link"
1287 * for the common case.
1288 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001289static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001290{
1291 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1292 if (likely(inode->i_op->follow_link))
1293 return follow;
1294
1295 /* This gets set once for the inode lifetime */
1296 spin_lock(&inode->i_lock);
1297 inode->i_opflags |= IOP_NOFOLLOW;
1298 spin_unlock(&inode->i_lock);
1299 }
1300 return 0;
1301}
1302
Al Viroce57dfc2011-03-13 19:58:58 -04001303static inline int walk_component(struct nameidata *nd, struct path *path,
1304 struct qstr *name, int type, int follow)
1305{
1306 struct inode *inode;
1307 int err;
1308 /*
1309 * "." and ".." are special - ".." especially so because it has
1310 * to be able to know about the current root directory and
1311 * parent relationships.
1312 */
1313 if (unlikely(type != LAST_NORM))
1314 return handle_dots(nd, type);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001315 err = lookup_fast(nd, name, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001316 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001317 if (err < 0)
1318 goto out_err;
1319
1320 err = lookup_slow(nd, name, path);
1321 if (err < 0)
1322 goto out_err;
1323
1324 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001325 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001326 err = -ENOENT;
1327 if (!inode)
1328 goto out_path_put;
1329
Linus Torvalds7813b942011-08-07 09:53:20 -07001330 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001331 if (nd->flags & LOOKUP_RCU) {
1332 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001333 err = -ECHILD;
1334 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001335 }
1336 }
Al Viroce57dfc2011-03-13 19:58:58 -04001337 BUG_ON(inode != path->dentry->d_inode);
1338 return 1;
1339 }
1340 path_to_nameidata(path, nd);
1341 nd->inode = inode;
1342 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001343
1344out_path_put:
1345 path_to_nameidata(path, nd);
1346out_err:
1347 terminate_walk(nd);
1348 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001349}
1350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351/*
Al Virob3563792011-03-14 21:54:55 -04001352 * This limits recursive symlink follows to 8, while
1353 * limiting consecutive symlinks to 40.
1354 *
1355 * Without that kind of total limit, nasty chains of consecutive
1356 * symlinks can cause almost arbitrarily long lookups.
1357 */
1358static inline int nested_symlink(struct path *path, struct nameidata *nd)
1359{
1360 int res;
1361
Al Virob3563792011-03-14 21:54:55 -04001362 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1363 path_put_conditional(path, nd);
1364 path_put(&nd->path);
1365 return -ELOOP;
1366 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001367 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001368
1369 nd->depth++;
1370 current->link_count++;
1371
1372 do {
1373 struct path link = *path;
1374 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001375
1376 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001377 if (res)
1378 break;
1379 res = walk_component(nd, path, &nd->last,
1380 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001381 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001382 } while (res > 0);
1383
1384 current->link_count--;
1385 nd->depth--;
1386 return res;
1387}
1388
1389/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001390 * We really don't want to look at inode->i_op->lookup
1391 * when we don't have to. So we keep a cache bit in
1392 * the inode ->i_opflags field that says "yes, we can
1393 * do lookup on this inode".
1394 */
1395static inline int can_lookup(struct inode *inode)
1396{
1397 if (likely(inode->i_opflags & IOP_LOOKUP))
1398 return 1;
1399 if (likely(!inode->i_op->lookup))
1400 return 0;
1401
1402 /* We do this once for the lifetime of the inode */
1403 spin_lock(&inode->i_lock);
1404 inode->i_opflags |= IOP_LOOKUP;
1405 spin_unlock(&inode->i_lock);
1406 return 1;
1407}
1408
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001409/*
1410 * We can do the critical dentry name comparison and hashing
1411 * operations one word at a time, but we are limited to:
1412 *
1413 * - Architectures with fast unaligned word accesses. We could
1414 * do a "get_unaligned()" if this helps and is sufficiently
1415 * fast.
1416 *
1417 * - Little-endian machines (so that we can generate the mask
1418 * of low bytes efficiently). Again, we *could* do a byte
1419 * swapping load on big-endian architectures if that is not
1420 * expensive enough to make the optimization worthless.
1421 *
1422 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1423 * do not trap on the (extremely unlikely) case of a page
1424 * crossing operation.
1425 *
1426 * - Furthermore, we need an efficient 64-bit compile for the
1427 * 64-bit case in order to generate the "number of bytes in
1428 * the final mask". Again, that could be replaced with a
1429 * efficient population count instruction or similar.
1430 */
1431#ifdef CONFIG_DCACHE_WORD_ACCESS
1432
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001433#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001434
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001435#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001436
1437static inline unsigned int fold_hash(unsigned long hash)
1438{
1439 hash += hash >> (8*sizeof(int));
1440 return hash;
1441}
1442
1443#else /* 32-bit case */
1444
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001445#define fold_hash(x) (x)
1446
1447#endif
1448
1449unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1450{
1451 unsigned long a, mask;
1452 unsigned long hash = 0;
1453
1454 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001455 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001456 if (len < sizeof(unsigned long))
1457 break;
1458 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001459 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001460 name += sizeof(unsigned long);
1461 len -= sizeof(unsigned long);
1462 if (!len)
1463 goto done;
1464 }
1465 mask = ~(~0ul << len*8);
1466 hash += mask & a;
1467done:
1468 return fold_hash(hash);
1469}
1470EXPORT_SYMBOL(full_name_hash);
1471
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001472/*
1473 * Calculate the length and hash of the path component, and
1474 * return the length of the component;
1475 */
1476static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1477{
Linus Torvalds36126f82012-05-26 10:43:17 -07001478 unsigned long a, b, adata, bdata, mask, hash, len;
1479 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001480
1481 hash = a = 0;
1482 len = -sizeof(unsigned long);
1483 do {
1484 hash = (hash + a) * 9;
1485 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001486 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001487 b = a ^ REPEAT_BYTE('/');
1488 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001489
Linus Torvalds36126f82012-05-26 10:43:17 -07001490 adata = prep_zero_mask(a, adata, &constants);
1491 bdata = prep_zero_mask(b, bdata, &constants);
1492
1493 mask = create_zero_mask(adata | bdata);
1494
1495 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001496 *hashp = fold_hash(hash);
1497
Linus Torvalds36126f82012-05-26 10:43:17 -07001498 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001499}
1500
1501#else
1502
Linus Torvalds0145acc2012-03-02 14:32:59 -08001503unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1504{
1505 unsigned long hash = init_name_hash();
1506 while (len--)
1507 hash = partial_name_hash(*name++, hash);
1508 return end_name_hash(hash);
1509}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001510EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001511
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001512/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001513 * We know there's a real path component here of at least
1514 * one character.
1515 */
1516static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1517{
1518 unsigned long hash = init_name_hash();
1519 unsigned long len = 0, c;
1520
1521 c = (unsigned char)*name;
1522 do {
1523 len++;
1524 hash = partial_name_hash(c, hash);
1525 c = (unsigned char)name[len];
1526 } while (c && c != '/');
1527 *hashp = end_name_hash(hash);
1528 return len;
1529}
1530
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001531#endif
1532
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001533/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001535 * This is the basic name resolution function, turning a pathname into
1536 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001538 * Returns 0 and nd will have valid dentry and mnt on success.
1539 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 */
Al Viro6de88d72009-08-09 01:41:57 +04001541static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542{
1543 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
1546 while (*name=='/')
1547 name++;
1548 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001549 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 /* At this point we know we have a real path component. */
1552 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001554 long len;
Al Virofe479a52011-02-22 15:10:03 -05001555 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Al Viro52094c82011-02-21 21:34:47 -05001557 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (err)
1559 break;
1560
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001561 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001563 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Al Virofe479a52011-02-22 15:10:03 -05001565 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001566 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001567 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001568 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001569 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001570 nd->flags |= LOOKUP_JUMPED;
1571 }
Al Virofe479a52011-02-22 15:10:03 -05001572 break;
1573 case 1:
1574 type = LAST_DOT;
1575 }
Al Viro5a202bc2011-03-08 14:17:44 -05001576 if (likely(type == LAST_NORM)) {
1577 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001578 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001579 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1580 err = parent->d_op->d_hash(parent, nd->inode,
1581 &this);
1582 if (err < 0)
1583 break;
1584 }
1585 }
Al Virofe479a52011-02-22 15:10:03 -05001586
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001587 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001589 /*
1590 * If it wasn't NUL, we know it was '/'. Skip that
1591 * slash, and continue until no more slashes.
1592 */
1593 do {
1594 len++;
1595 } while (unlikely(name[len] == '/'));
1596 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001597 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001598 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
Al Viroce57dfc2011-03-13 19:58:58 -04001600 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1601 if (err < 0)
1602 return err;
Al Virofe479a52011-02-22 15:10:03 -05001603
Al Viroce57dfc2011-03-13 19:58:58 -04001604 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001605 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001607 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001608 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001609 if (can_lookup(nd->inode))
1610 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001612 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 /* here ends the main loop */
1614
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615last_component:
Al Virob3563792011-03-14 21:54:55 -04001616 nd->last = this;
1617 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001618 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 }
Al Viro951361f2011-03-04 14:44:37 -05001620 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 return err;
1622}
1623
Al Viro70e9b352011-03-05 21:12:22 -05001624static int path_init(int dfd, const char *name, unsigned int flags,
1625 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001627 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001628 int fput_needed;
1629 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001632 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001634 if (flags & LOOKUP_ROOT) {
1635 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001636 if (*name) {
1637 if (!inode->i_op->lookup)
1638 return -ENOTDIR;
1639 retval = inode_permission(inode, MAY_EXEC);
1640 if (retval)
1641 return retval;
1642 }
Al Viro5b6ca022011-03-09 23:04:47 -05001643 nd->path = nd->root;
1644 nd->inode = inode;
1645 if (flags & LOOKUP_RCU) {
Andi Kleen962830d2012-05-08 13:32:02 +09301646 br_read_lock(&vfsmount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001647 rcu_read_lock();
1648 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1649 } else {
1650 path_get(&nd->path);
1651 }
1652 return 0;
1653 }
1654
Al Viro2a737872009-04-07 11:49:53 -04001655 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001658 if (flags & LOOKUP_RCU) {
Andi Kleen962830d2012-05-08 13:32:02 +09301659 br_read_lock(&vfsmount_lock);
Al Viroe41f7d42011-02-22 14:02:58 -05001660 rcu_read_lock();
1661 set_root_rcu(nd);
1662 } else {
1663 set_root(nd);
1664 path_get(&nd->root);
1665 }
Al Viro2a737872009-04-07 11:49:53 -04001666 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001667 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001668 if (flags & LOOKUP_RCU) {
1669 struct fs_struct *fs = current->fs;
1670 unsigned seq;
1671
Andi Kleen962830d2012-05-08 13:32:02 +09301672 br_read_lock(&vfsmount_lock);
Al Viroe41f7d42011-02-22 14:02:58 -05001673 rcu_read_lock();
1674
1675 do {
1676 seq = read_seqcount_begin(&fs->seq);
1677 nd->path = fs->pwd;
1678 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1679 } while (read_seqcount_retry(&fs->seq, seq));
1680 } else {
1681 get_fs_pwd(current->fs, &nd->path);
1682 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001683 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001684 struct dentry *dentry;
1685
Al Viro1abf0c72011-03-13 03:51:11 -04001686 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001687 retval = -EBADF;
1688 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001689 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001690
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001691 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001692
Al Virof52e0c12011-03-14 18:56:51 -04001693 if (*name) {
1694 retval = -ENOTDIR;
1695 if (!S_ISDIR(dentry->d_inode->i_mode))
1696 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001697
Al Viro4ad5abb2011-06-20 19:57:03 -04001698 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001699 if (retval)
1700 goto fput_fail;
1701 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001702
Jan Blunck5dd784d02008-02-14 19:34:38 -08001703 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001704 if (flags & LOOKUP_RCU) {
1705 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001706 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001707 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Andi Kleen962830d2012-05-08 13:32:02 +09301708 br_read_lock(&vfsmount_lock);
Al Viroe41f7d42011-02-22 14:02:58 -05001709 rcu_read_lock();
1710 } else {
1711 path_get(&file->f_path);
1712 fput_light(file, fput_needed);
1713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 }
Al Viroe41f7d42011-02-22 14:02:58 -05001715
Nick Piggin31e6b012011-01-07 17:49:52 +11001716 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001717 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001718
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001719fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001720 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001721out_fail:
1722 return retval;
1723}
1724
Al Virobd92d7f2011-03-14 19:54:59 -04001725static inline int lookup_last(struct nameidata *nd, struct path *path)
1726{
1727 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1728 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1729
1730 nd->flags &= ~LOOKUP_PARENT;
1731 return walk_component(nd, path, &nd->last, nd->last_type,
1732 nd->flags & LOOKUP_FOLLOW);
1733}
1734
Al Viro9b4a9b12009-04-07 11:44:16 -04001735/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001736static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001737 unsigned int flags, struct nameidata *nd)
1738{
Al Viro70e9b352011-03-05 21:12:22 -05001739 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001740 struct path path;
1741 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001742
1743 /*
1744 * Path walking is largely split up into 2 different synchronisation
1745 * schemes, rcu-walk and ref-walk (explained in
1746 * Documentation/filesystems/path-lookup.txt). These share much of the
1747 * path walk code, but some things particularly setup, cleanup, and
1748 * following mounts are sufficiently divergent that functions are
1749 * duplicated. Typically there is a function foo(), and its RCU
1750 * analogue, foo_rcu().
1751 *
1752 * -ECHILD is the error number of choice (just to avoid clashes) that
1753 * is returned if some aspect of an rcu-walk fails. Such an error must
1754 * be handled by restarting a traditional ref-walk (which will always
1755 * be able to complete).
1756 */
Al Virobd92d7f2011-03-14 19:54:59 -04001757 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001758
Al Virobd92d7f2011-03-14 19:54:59 -04001759 if (unlikely(err))
1760 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001761
1762 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001763 err = link_path_walk(name, nd);
1764
1765 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001766 err = lookup_last(nd, &path);
1767 while (err > 0) {
1768 void *cookie;
1769 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001770 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001771 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001772 if (err)
1773 break;
1774 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001775 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001776 }
1777 }
Al Viroee0827c2011-02-21 23:38:09 -05001778
Al Viro9f1fafe2011-03-25 11:00:12 -04001779 if (!err)
1780 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001781
1782 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1783 if (!nd->inode->i_op->lookup) {
1784 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001785 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001786 }
1787 }
Al Viro16c2cd72011-02-22 15:50:10 -05001788
Al Viro70e9b352011-03-05 21:12:22 -05001789 if (base)
1790 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001791
Al Viro5b6ca022011-03-09 23:04:47 -05001792 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001793 path_put(&nd->root);
1794 nd->root.mnt = NULL;
1795 }
Al Virobd92d7f2011-03-14 19:54:59 -04001796 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001797}
Nick Piggin31e6b012011-01-07 17:49:52 +11001798
Al Viroee0827c2011-02-21 23:38:09 -05001799static int do_path_lookup(int dfd, const char *name,
1800 unsigned int flags, struct nameidata *nd)
1801{
1802 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1803 if (unlikely(retval == -ECHILD))
1804 retval = path_lookupat(dfd, name, flags, nd);
1805 if (unlikely(retval == -ESTALE))
1806 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001807
1808 if (likely(!retval)) {
1809 if (unlikely(!audit_dummy_context())) {
1810 if (nd->path.dentry && nd->inode)
1811 audit_inode(name, nd->path.dentry);
1812 }
1813 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001814 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815}
1816
Al Viroc9c6cac2011-02-16 15:15:47 -05001817int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001818{
Al Viroc9c6cac2011-02-16 15:15:47 -05001819 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001820}
1821
Al Virod1811462008-08-02 00:49:18 -04001822int kern_path(const char *name, unsigned int flags, struct path *path)
1823{
1824 struct nameidata nd;
1825 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1826 if (!res)
1827 *path = nd.path;
1828 return res;
1829}
1830
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001831/**
1832 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1833 * @dentry: pointer to dentry of the base directory
1834 * @mnt: pointer to vfs mount of the base directory
1835 * @name: pointer to file name
1836 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001837 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001838 */
1839int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1840 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001841 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001842{
Al Viroe0a01242011-06-27 17:00:37 -04001843 struct nameidata nd;
1844 int err;
1845 nd.root.dentry = dentry;
1846 nd.root.mnt = mnt;
1847 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001848 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001849 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1850 if (!err)
1851 *path = nd.path;
1852 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001853}
1854
James Morris057f6c02007-04-26 00:12:05 -07001855/*
1856 * Restricted form of lookup. Doesn't follow links, single-component only,
1857 * needs parent already locked. Doesn't follow mounts.
1858 * SMP-safe.
1859 */
Adrian Bunka244e162006-03-31 02:32:11 -08001860static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
Jan Blunck4ac91372008-02-14 19:34:32 -08001862 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
1864
Christoph Hellwigeead1912007-10-16 23:25:38 -07001865/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001866 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001867 * @name: pathname component to lookup
1868 * @base: base directory to lookup from
1869 * @len: maximum length @len should be interpreted to
1870 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001871 * Note that this routine is purely a helper for filesystem usage and should
1872 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001873 * nameidata argument is passed to the filesystem methods and a filesystem
1874 * using this helper needs to be prepared for that.
1875 */
James Morris057f6c02007-04-26 00:12:05 -07001876struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1877{
James Morris057f6c02007-04-26 00:12:05 -07001878 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001879 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02001880 int err;
James Morris057f6c02007-04-26 00:12:05 -07001881
David Woodhouse2f9092e2009-04-20 23:18:37 +01001882 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1883
Al Viro6a96ba52011-03-07 23:49:20 -05001884 this.name = name;
1885 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08001886 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05001887 if (!len)
1888 return ERR_PTR(-EACCES);
1889
Al Viro6a96ba52011-03-07 23:49:20 -05001890 while (len--) {
1891 c = *(const unsigned char *)name++;
1892 if (c == '/' || c == '\0')
1893 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05001894 }
Al Viro5a202bc2011-03-08 14:17:44 -05001895 /*
1896 * See if the low-level filesystem might want
1897 * to use its own hash..
1898 */
1899 if (base->d_flags & DCACHE_OP_HASH) {
1900 int err = base->d_op->d_hash(base, base->d_inode, &this);
1901 if (err < 0)
1902 return ERR_PTR(err);
1903 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001904
Miklos Szeredicda309d2012-03-26 12:54:21 +02001905 err = inode_permission(base->d_inode, MAY_EXEC);
1906 if (err)
1907 return ERR_PTR(err);
1908
Christoph Hellwig49705b72005-11-08 21:35:06 -08001909 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001910}
1911
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001912int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1913 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914{
Al Viro2d8f3032008-07-22 09:59:21 -04001915 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001916 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001919
1920 BUG_ON(flags & LOOKUP_PARENT);
1921
1922 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001924 if (!err)
1925 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 }
1927 return err;
1928}
1929
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001930int user_path_at(int dfd, const char __user *name, unsigned flags,
1931 struct path *path)
1932{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07001933 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001934}
1935
Al Viro2ad94ae2008-07-21 09:32:51 -04001936static int user_path_parent(int dfd, const char __user *path,
1937 struct nameidata *nd, char **name)
1938{
1939 char *s = getname(path);
1940 int error;
1941
1942 if (IS_ERR(s))
1943 return PTR_ERR(s);
1944
1945 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1946 if (error)
1947 putname(s);
1948 else
1949 *name = s;
1950
1951 return error;
1952}
1953
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954/*
1955 * It's inline, so penalty for filesystems that don't use sticky bit is
1956 * minimal.
1957 */
1958static inline int check_sticky(struct inode *dir, struct inode *inode)
1959{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001960 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11001961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 if (!(dir->i_mode & S_ISVTX))
1963 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001964 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001966 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08001968 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
1970
1971/*
1972 * Check whether we can remove a link victim from directory dir, check
1973 * whether the type of victim is right.
1974 * 1. We can't do it if dir is read-only (done in permission())
1975 * 2. We should have write and exec permissions on dir
1976 * 3. We can't remove anything from append-only dir
1977 * 4. We can't do anything with immutable dir (done in permission())
1978 * 5. If the sticky bit on dir is set we should either
1979 * a. be owner of dir, or
1980 * b. be owner of victim, or
1981 * c. have CAP_FOWNER capability
1982 * 6. If the victim is append-only or immutable we can't do antyhing with
1983 * links pointing to it.
1984 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1985 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1986 * 9. We can't remove a root or mountpoint.
1987 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1988 * nfs_async_unlink().
1989 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001990static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
1992 int error;
1993
1994 if (!victim->d_inode)
1995 return -ENOENT;
1996
1997 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001998 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Al Virof419a2e2008-07-22 00:07:17 -04002000 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (error)
2002 return error;
2003 if (IS_APPEND(dir))
2004 return -EPERM;
2005 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002006 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 return -EPERM;
2008 if (isdir) {
2009 if (!S_ISDIR(victim->d_inode->i_mode))
2010 return -ENOTDIR;
2011 if (IS_ROOT(victim))
2012 return -EBUSY;
2013 } else if (S_ISDIR(victim->d_inode->i_mode))
2014 return -EISDIR;
2015 if (IS_DEADDIR(dir))
2016 return -ENOENT;
2017 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2018 return -EBUSY;
2019 return 0;
2020}
2021
2022/* Check whether we can create an object with dentry child in directory
2023 * dir.
2024 * 1. We can't do it if child already exists (open has special treatment for
2025 * this case, but since we are inlined it's OK)
2026 * 2. We can't do it if dir is read-only (done in permission())
2027 * 3. We should have write and exec permissions on dir
2028 * 4. We can't do it if dir is immutable (done in permission())
2029 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002030static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031{
2032 if (child->d_inode)
2033 return -EEXIST;
2034 if (IS_DEADDIR(dir))
2035 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002036 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037}
2038
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039/*
2040 * p1 and p2 should be directories on the same fs.
2041 */
2042struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2043{
2044 struct dentry *p;
2045
2046 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002047 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return NULL;
2049 }
2050
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002051 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002053 p = d_ancestor(p2, p1);
2054 if (p) {
2055 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2056 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2057 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 }
2059
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002060 p = d_ancestor(p1, p2);
2061 if (p) {
2062 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2063 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2064 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
2066
Ingo Molnarf2eace22006-07-03 00:25:05 -07002067 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2068 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 return NULL;
2070}
2071
2072void unlock_rename(struct dentry *p1, struct dentry *p2)
2073{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002074 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002076 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002077 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 }
2079}
2080
Al Viro4acdaf22011-07-26 01:42:34 -04002081int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 struct nameidata *nd)
2083{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002084 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 if (error)
2087 return error;
2088
Al Viroacfa4382008-12-04 10:06:33 -05002089 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 return -EACCES; /* shouldn't it be ENOSYS? */
2091 mode &= S_IALLUGO;
2092 mode |= S_IFREG;
2093 error = security_inode_create(dir, dentry, mode);
2094 if (error)
2095 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002097 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002098 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 return error;
2100}
2101
Al Viro73d049a2011-03-11 12:08:24 -05002102static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002104 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 struct inode *inode = dentry->d_inode;
2106 int error;
2107
Al Virobcda7652011-03-13 16:42:14 -04002108 /* O_PATH? */
2109 if (!acc_mode)
2110 return 0;
2111
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 if (!inode)
2113 return -ENOENT;
2114
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002115 switch (inode->i_mode & S_IFMT) {
2116 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002118 case S_IFDIR:
2119 if (acc_mode & MAY_WRITE)
2120 return -EISDIR;
2121 break;
2122 case S_IFBLK:
2123 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002124 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002126 /*FALLTHRU*/
2127 case S_IFIFO:
2128 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002130 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002131 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002132
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002133 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002134 if (error)
2135 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002136
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 /*
2138 * An append-only file must be opened in append mode for writing.
2139 */
2140 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002141 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002142 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002144 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 }
2146
2147 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002148 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002149 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002151 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002152}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
Jeff Laytone1181ee2010-12-07 16:19:50 -05002154static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002155{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002156 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002157 struct inode *inode = path->dentry->d_inode;
2158 int error = get_write_access(inode);
2159 if (error)
2160 return error;
2161 /*
2162 * Refuse to truncate files with mandatory locks held on them.
2163 */
2164 error = locks_verify_locked(inode);
2165 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002166 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002167 if (!error) {
2168 error = do_truncate(path->dentry, 0,
2169 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002170 filp);
Al Viro7715b522009-12-16 03:54:00 -05002171 }
2172 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002173 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174}
2175
Dave Hansend57999e2008-02-15 14:37:27 -08002176static inline int open_to_namei_flags(int flag)
2177{
Al Viro8a5e9292011-06-25 19:15:54 -04002178 if ((flag & O_ACCMODE) == 3)
2179 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002180 return flag;
2181}
2182
Miklos Szeredid18e9002012-06-05 15:10:17 +02002183static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2184{
2185 int error = security_path_mknod(dir, dentry, mode, 0);
2186 if (error)
2187 return error;
2188
2189 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2190 if (error)
2191 return error;
2192
2193 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2194}
2195
2196static struct file *atomic_open(struct nameidata *nd, struct dentry *dentry,
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002197 struct path *path, struct opendata *od,
2198 const struct open_flags *op,
Miklos Szeredid18e9002012-06-05 15:10:17 +02002199 int *want_write, bool need_lookup,
2200 bool *created)
2201{
2202 struct inode *dir = nd->path.dentry->d_inode;
2203 unsigned open_flag = open_to_namei_flags(op->open_flag);
2204 umode_t mode;
2205 int error;
2206 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002207 struct file *filp;
2208 int create_error = 0;
2209 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2210
2211 BUG_ON(dentry->d_inode);
2212
2213 /* Don't create child dentry for a dead directory. */
2214 if (unlikely(IS_DEADDIR(dir))) {
2215 filp = ERR_PTR(-ENOENT);
2216 goto out;
2217 }
2218
2219 mode = op->mode & S_IALLUGO;
2220 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2221 mode &= ~current_umask();
2222
2223 if (open_flag & O_EXCL) {
2224 open_flag &= ~O_TRUNC;
2225 *created = true;
2226 }
2227
2228 /*
2229 * Checking write permission is tricky, bacuse we don't know if we are
2230 * going to actually need it: O_CREAT opens should work as long as the
2231 * file exists. But checking existence breaks atomicity. The trick is
2232 * to check access and if not granted clear O_CREAT from the flags.
2233 *
2234 * Another problem is returing the "right" error value (e.g. for an
2235 * O_EXCL open we want to return EEXIST not EROFS).
2236 */
2237 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2238 (open_flag & O_ACCMODE) != O_RDONLY) {
2239 error = mnt_want_write(nd->path.mnt);
2240 if (!error) {
2241 *want_write = 1;
2242 } else if (!(open_flag & O_CREAT)) {
2243 /*
2244 * No O_CREATE -> atomicity not a requirement -> fall
2245 * back to lookup + open
2246 */
2247 goto no_open;
2248 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2249 /* Fall back and fail with the right error */
2250 create_error = error;
2251 goto no_open;
2252 } else {
2253 /* No side effects, safe to clear O_CREAT */
2254 create_error = error;
2255 open_flag &= ~O_CREAT;
2256 }
2257 }
2258
2259 if (open_flag & O_CREAT) {
2260 error = may_o_create(&nd->path, dentry, op->mode);
2261 if (error) {
2262 create_error = error;
2263 if (open_flag & O_EXCL)
2264 goto no_open;
2265 open_flag &= ~O_CREAT;
2266 }
2267 }
2268
2269 if (nd->flags & LOOKUP_DIRECTORY)
2270 open_flag |= O_DIRECTORY;
2271
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002272 od->dentry = DENTRY_NOT_SET;
2273 od->mnt = nd->path.mnt;
2274 filp = dir->i_op->atomic_open(dir, dentry, od, open_flag, mode,
Miklos Szeredid18e9002012-06-05 15:10:17 +02002275 created);
2276 if (IS_ERR(filp)) {
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002277 if (WARN_ON(od->dentry != DENTRY_NOT_SET))
2278 dput(od->dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002279
2280 if (create_error && PTR_ERR(filp) == -ENOENT)
2281 filp = ERR_PTR(create_error);
2282 goto out;
2283 }
2284
2285 acc_mode = op->acc_mode;
2286 if (*created) {
2287 fsnotify_create(dir, dentry);
2288 acc_mode = MAY_OPEN;
2289 }
2290
2291 if (!filp) {
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002292 if (WARN_ON(od->dentry == DENTRY_NOT_SET)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002293 filp = ERR_PTR(-EIO);
2294 goto out;
2295 }
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002296 if (od->dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002297 dput(dentry);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002298 dentry = od->dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002299 }
2300 goto looked_up;
2301 }
2302
2303 /*
2304 * We didn't have the inode before the open, so check open permission
2305 * here.
2306 */
2307 error = may_open(&filp->f_path, acc_mode, open_flag);
2308 if (error)
2309 goto out_fput;
2310
2311 error = open_check_o_direct(filp);
2312 if (error)
2313 goto out_fput;
2314
2315out:
2316 dput(dentry);
2317 return filp;
2318
2319out_fput:
2320 fput(filp);
2321 filp = ERR_PTR(error);
2322 goto out;
2323
2324no_open:
2325 if (need_lookup) {
2326 dentry = lookup_real(dir, dentry, nd);
2327 if (IS_ERR(dentry))
2328 return ERR_CAST(dentry);
2329
2330 if (create_error) {
2331 int open_flag = op->open_flag;
2332
2333 filp = ERR_PTR(create_error);
2334 if ((open_flag & O_EXCL)) {
2335 if (!dentry->d_inode)
2336 goto out;
2337 } else if (!dentry->d_inode) {
2338 goto out;
2339 } else if ((open_flag & O_TRUNC) &&
2340 S_ISREG(dentry->d_inode->i_mode)) {
2341 goto out;
2342 }
2343 /* will fail later, go on to get the right error */
2344 }
2345 }
2346looked_up:
2347 path->dentry = dentry;
2348 path->mnt = nd->path.mnt;
2349 return NULL;
2350}
2351
Nick Piggin31e6b012011-01-07 17:49:52 +11002352/*
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002353 * Lookup, maybe create and open the last component
2354 *
2355 * Must be called with i_mutex held on parent.
2356 *
2357 * Returns open file or NULL on success, error otherwise. NULL means no open
2358 * was performed, only lookup.
2359 */
2360static struct file *lookup_open(struct nameidata *nd, struct path *path,
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002361 struct opendata *od,
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002362 const struct open_flags *op,
2363 int *want_write, bool *created)
2364{
2365 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002366 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002367 struct dentry *dentry;
2368 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002369 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002370
2371 *created = false;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002372 dentry = lookup_dcache(&nd->last, dir, nd, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002373 if (IS_ERR(dentry))
2374 return ERR_CAST(dentry);
2375
Miklos Szeredid18e9002012-06-05 15:10:17 +02002376 /* Cached positive dentry: will open in f_op->open */
2377 if (!need_lookup && dentry->d_inode)
2378 goto out_no_open;
2379
2380 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002381 return atomic_open(nd, dentry, path, od, op, want_write,
Miklos Szeredid18e9002012-06-05 15:10:17 +02002382 need_lookup, created);
2383 }
2384
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002385 if (need_lookup) {
2386 BUG_ON(dentry->d_inode);
2387
2388 dentry = lookup_real(dir_inode, dentry, nd);
2389 if (IS_ERR(dentry))
2390 return ERR_CAST(dentry);
2391 }
2392
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002393 /* Negative dentry, just create the file */
2394 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2395 umode_t mode = op->mode;
2396 if (!IS_POSIXACL(dir->d_inode))
2397 mode &= ~current_umask();
2398 /*
2399 * This write is needed to ensure that a
2400 * rw->ro transition does not occur between
2401 * the time when the file is created and when
2402 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002403 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002404 */
2405 error = mnt_want_write(nd->path.mnt);
2406 if (error)
2407 goto out_dput;
2408 *want_write = 1;
2409 *created = true;
2410 error = security_path_mknod(&nd->path, dentry, mode, 0);
2411 if (error)
2412 goto out_dput;
2413 error = vfs_create(dir->d_inode, dentry, mode, nd);
2414 if (error)
2415 goto out_dput;
2416 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002417out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002418 path->dentry = dentry;
2419 path->mnt = nd->path.mnt;
2420 return NULL;
2421
2422out_dput:
2423 dput(dentry);
2424 return ERR_PTR(error);
2425}
2426
2427/*
Al Virofe2d35f2011-03-05 22:58:25 -05002428 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002429 */
Al Virofb1cc552009-12-24 01:58:28 -05002430static struct file *do_last(struct nameidata *nd, struct path *path,
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002431 struct opendata *od, const struct open_flags *op,
2432 const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002433{
Al Viroa1e28032009-12-24 02:12:06 -05002434 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002435 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002436 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002437 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002438 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002439 struct file *filp;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002440 struct inode *inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002441 bool created;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002442 int symlink_ok = 0;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002443 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2444 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002445 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002446
Al Viroc3e380b2011-02-23 13:39:45 -05002447 nd->flags &= ~LOOKUP_PARENT;
2448 nd->flags |= op->intent;
2449
Al Viro1f36f772009-12-26 10:56:19 -05002450 switch (nd->last_type) {
2451 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002452 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002453 error = handle_dots(nd, nd->last_type);
2454 if (error)
2455 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002456 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002457 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002458 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002459 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002460 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002461 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002462 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002463 error = -EISDIR;
2464 goto exit;
2465 }
2466 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002467 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002468 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002469 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002470 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002471 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002472 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002473 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002474
Al Viroca344a892011-03-09 00:36:45 -05002475 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002476 if (nd->last.name[nd->last.len])
2477 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002478 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2479 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002480 /* we _can_ be in RCU mode here */
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002481 error = lookup_fast(nd, &nd->last, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002482 if (likely(!error))
2483 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002484
Miklos Szeredi71574862012-06-05 15:10:14 +02002485 if (error < 0)
2486 goto exit;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002487
Miklos Szeredi71574862012-06-05 15:10:14 +02002488 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002489 } else {
2490 /* create side of things */
2491 /*
2492 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2493 * has been cleared when we got to the last component we are
2494 * about to look up
2495 */
2496 error = complete_walk(nd);
2497 if (error)
2498 return ERR_PTR(error);
2499
2500 audit_inode(pathname, dir);
2501 error = -EISDIR;
2502 /* trailing slashes? */
2503 if (nd->last.name[nd->last.len])
2504 goto exit;
Al Virofe2d35f2011-03-05 22:58:25 -05002505 }
2506
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002507retry_lookup:
Al Viroa1e28032009-12-24 02:12:06 -05002508 mutex_lock(&dir->d_inode->i_mutex);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002509 filp = lookup_open(nd, path, od, op, &want_write, &created);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002510 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002511
Miklos Szeredid18e9002012-06-05 15:10:17 +02002512 if (filp) {
2513 if (IS_ERR(filp))
2514 goto out;
2515
2516 if (created || !S_ISREG(filp->f_path.dentry->d_inode->i_mode))
2517 will_truncate = 0;
2518
2519 audit_inode(pathname, filp->f_path.dentry);
2520 goto opened;
2521 }
Al Virofb1cc552009-12-24 01:58:28 -05002522
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002523 if (created) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002524 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002525 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002526 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002527 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002528 path_to_nameidata(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05002529 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002530 }
2531
2532 /*
2533 * It already exists.
2534 */
Al Virofb1cc552009-12-24 01:58:28 -05002535 audit_inode(pathname, path->dentry);
2536
Miklos Szeredid18e9002012-06-05 15:10:17 +02002537 /*
2538 * If atomic_open() acquired write access it is dropped now due to
2539 * possible mount and symlink following (this might be optimized away if
2540 * necessary...)
2541 */
2542 if (want_write) {
2543 mnt_drop_write(nd->path.mnt);
2544 want_write = 0;
2545 }
2546
Al Virofb1cc552009-12-24 01:58:28 -05002547 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002548 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002549 goto exit_dput;
2550
David Howells9875cf82011-01-14 18:45:21 +00002551 error = follow_managed(path, nd->flags);
2552 if (error < 0)
2553 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002554
Al Viroa3fbbde2011-11-07 21:21:26 +00002555 if (error)
2556 nd->flags |= LOOKUP_JUMPED;
2557
Miklos Szeredidecf3402012-05-21 17:30:08 +02002558 BUG_ON(nd->flags & LOOKUP_RCU);
2559 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002560finish_lookup:
2561 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002562 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002563 if (!inode) {
2564 path_to_nameidata(path, nd);
2565 goto exit;
2566 }
Al Viro9e67f362009-12-26 07:04:50 -05002567
Miklos Szeredid45ea862012-05-21 17:30:09 +02002568 if (should_follow_link(inode, !symlink_ok)) {
2569 if (nd->flags & LOOKUP_RCU) {
2570 if (unlikely(unlazy_walk(nd, path->dentry))) {
2571 error = -ECHILD;
2572 goto exit;
2573 }
2574 }
2575 BUG_ON(inode != path->dentry->d_inode);
Al Virofb1cc552009-12-24 01:58:28 -05002576 return NULL;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002577 }
Al Virofb1cc552009-12-24 01:58:28 -05002578
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002579 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2580 path_to_nameidata(path, nd);
2581 } else {
2582 save_parent.dentry = nd->path.dentry;
2583 save_parent.mnt = mntget(path->mnt);
2584 nd->path.dentry = path->dentry;
2585
2586 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002587 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002588 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2589 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002590 if (error) {
2591 path_put(&save_parent);
Miklos Szeredi097b1802012-03-06 13:56:33 +01002592 return ERR_PTR(error);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002593 }
Al Virofb1cc552009-12-24 01:58:28 -05002594 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002595 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002596 goto exit;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002597 error = -ENOTDIR;
2598 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2599 goto exit;
Miklos Szeredid7fdd7f2012-05-21 17:30:13 +02002600 audit_inode(pathname, nd->path.dentry);
Al Viro67ee3ad2009-12-26 07:01:01 -05002601ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002602 if (!S_ISREG(nd->inode->i_mode))
2603 will_truncate = 0;
2604
Al Viro0f9d1a12011-03-09 00:13:14 -05002605 if (will_truncate) {
2606 error = mnt_want_write(nd->path.mnt);
2607 if (error)
2608 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002609 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002610 }
Al Viroca344a892011-03-09 00:36:45 -05002611common:
Al Virobcda7652011-03-13 16:42:14 -04002612 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002613 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002614 goto exit;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002615 od->mnt = nd->path.mnt;
2616 filp = finish_open(od, nd->path.dentry, NULL);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002617 if (filp == ERR_PTR(-EOPENSTALE) && save_parent.dentry && !retried) {
2618 BUG_ON(save_parent.dentry != dir);
2619 path_put(&nd->path);
2620 nd->path = save_parent;
2621 nd->inode = dir->d_inode;
2622 save_parent.mnt = NULL;
2623 save_parent.dentry = NULL;
2624 if (want_write) {
2625 mnt_drop_write(nd->path.mnt);
2626 want_write = 0;
2627 }
2628 retried = true;
2629 goto retry_lookup;
2630 }
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002631 if (IS_ERR(filp))
2632 goto out;
2633 error = open_check_o_direct(filp);
2634 if (error)
2635 goto exit_fput;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002636opened:
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002637 error = ima_file_check(filp, op->acc_mode);
2638 if (error)
2639 goto exit_fput;
2640
2641 if (will_truncate) {
2642 error = handle_truncate(filp);
2643 if (error)
2644 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05002645 }
Al Viroca344a892011-03-09 00:36:45 -05002646out:
2647 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002648 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002649 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002650 terminate_walk(nd);
Al Virofb1cc552009-12-24 01:58:28 -05002651 return filp;
2652
Al Virofb1cc552009-12-24 01:58:28 -05002653exit_dput:
2654 path_put_conditional(path, nd);
2655exit:
Al Viroca344a892011-03-09 00:36:45 -05002656 filp = ERR_PTR(error);
2657 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002658exit_fput:
2659 fput(filp);
2660 goto exit;
2661
Al Virofb1cc552009-12-24 01:58:28 -05002662}
2663
Al Viro13aab422011-02-23 17:54:08 -05002664static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002665 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666{
Al Virofe2d35f2011-03-05 22:58:25 -05002667 struct file *base = NULL;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002668 struct opendata od;
2669 struct file *res;
Al Viro9850c052010-01-13 15:01:15 -05002670 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002671 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002672
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002673 od.filp = get_empty_filp();
2674 if (!od.filp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002675 return ERR_PTR(-ENFILE);
2676
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002677 od.filp->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11002678
Al Viro73d049a2011-03-11 12:08:24 -05002679 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002680 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002681 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002682
Al Virofe2d35f2011-03-05 22:58:25 -05002683 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002684 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002685 if (unlikely(error))
2686 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002688 res = do_last(nd, &path, &od, op, pathname);
2689 while (unlikely(!res)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002690 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002691 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002692 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002693 path_put_conditional(&path, nd);
2694 path_put(&nd->path);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002695 res = ERR_PTR(-ELOOP);
Al Viro40b39132011-03-09 16:22:18 -05002696 break;
2697 }
Al Viro73d049a2011-03-11 12:08:24 -05002698 nd->flags |= LOOKUP_PARENT;
2699 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002700 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002701 if (unlikely(error))
Al Viro6d7b5aa2012-06-10 04:15:17 -04002702 goto out_filp;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002703 res = do_last(nd, &path, &od, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002704 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002705 }
Al Viro10fa8e62009-12-26 07:09:49 -05002706out:
Al Viro73d049a2011-03-11 12:08:24 -05002707 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2708 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002709 if (base)
2710 fput(base);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002711 if (od.filp) {
2712 BUG_ON(od.filp->f_path.dentry);
2713 put_filp(od.filp);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002714 }
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002715 if (res == ERR_PTR(-EOPENSTALE)) {
2716 if (flags & LOOKUP_RCU)
2717 res = ERR_PTR(-ECHILD);
2718 else
2719 res = ERR_PTR(-ESTALE);
2720 }
2721 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
Nick Piggin31e6b012011-01-07 17:49:52 +11002723out_filp:
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002724 res = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002725 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726}
2727
Al Viro13aab422011-02-23 17:54:08 -05002728struct file *do_filp_open(int dfd, const char *pathname,
2729 const struct open_flags *op, int flags)
2730{
Al Viro73d049a2011-03-11 12:08:24 -05002731 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002732 struct file *filp;
2733
Al Viro73d049a2011-03-11 12:08:24 -05002734 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002735 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002736 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002737 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002738 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002739 return filp;
2740}
2741
Al Viro73d049a2011-03-11 12:08:24 -05002742struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2743 const char *name, const struct open_flags *op, int flags)
2744{
2745 struct nameidata nd;
2746 struct file *file;
2747
2748 nd.root.mnt = mnt;
2749 nd.root.dentry = dentry;
2750
2751 flags |= LOOKUP_ROOT;
2752
Al Virobcda7652011-03-13 16:42:14 -04002753 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002754 return ERR_PTR(-ELOOP);
2755
2756 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2757 if (unlikely(file == ERR_PTR(-ECHILD)))
2758 file = path_openat(-1, name, &nd, op, flags);
2759 if (unlikely(file == ERR_PTR(-ESTALE)))
2760 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2761 return file;
2762}
2763
Al Viroed75e952011-06-27 16:53:43 -04002764struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002766 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002767 struct nameidata nd;
2768 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2769 if (error)
2770 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002772 /*
2773 * Yucky last component or no last component at all?
2774 * (foo/., foo/.., /////)
2775 */
Al Viroed75e952011-06-27 16:53:43 -04002776 if (nd.last_type != LAST_NORM)
2777 goto out;
2778 nd.flags &= ~LOOKUP_PARENT;
2779 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002780
2781 /*
2782 * Do the final lookup.
2783 */
Al Viroed75e952011-06-27 16:53:43 -04002784 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2785 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if (IS_ERR(dentry))
2787 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002788
Al Viroe9baf6e2008-05-15 04:49:12 -04002789 if (dentry->d_inode)
2790 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002791 /*
2792 * Special case - lookup gave negative, but... we had foo/bar/
2793 * From the vfs_mknod() POV we just have a negative dentry -
2794 * all is fine. Let's be bastards - you had / on the end, you've
2795 * been asking for (non-existent) directory. -ENOENT for you.
2796 */
Al Viroed75e952011-06-27 16:53:43 -04002797 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002798 dput(dentry);
2799 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002800 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002801 }
Al Viroed75e952011-06-27 16:53:43 -04002802 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002804eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002806 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807fail:
Al Viroed75e952011-06-27 16:53:43 -04002808 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2809out:
2810 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 return dentry;
2812}
Al Virodae6ad82011-06-26 11:50:15 -04002813EXPORT_SYMBOL(kern_path_create);
2814
2815struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2816{
2817 char *tmp = getname(pathname);
2818 struct dentry *res;
2819 if (IS_ERR(tmp))
2820 return ERR_CAST(tmp);
2821 res = kern_path_create(dfd, tmp, path, is_dir);
2822 putname(tmp);
2823 return res;
2824}
2825EXPORT_SYMBOL(user_path_create);
2826
Al Viro1a67aaf2011-07-26 01:52:52 -04002827int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002829 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830
2831 if (error)
2832 return error;
2833
Eric W. Biederman975d6b32011-11-13 12:16:43 -08002834 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 return -EPERM;
2836
Al Viroacfa4382008-12-04 10:06:33 -05002837 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 return -EPERM;
2839
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002840 error = devcgroup_inode_mknod(mode, dev);
2841 if (error)
2842 return error;
2843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 error = security_inode_mknod(dir, dentry, mode, dev);
2845 if (error)
2846 return error;
2847
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002849 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002850 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 return error;
2852}
2853
Al Virof69aac02011-07-26 04:31:40 -04002854static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08002855{
2856 switch (mode & S_IFMT) {
2857 case S_IFREG:
2858 case S_IFCHR:
2859 case S_IFBLK:
2860 case S_IFIFO:
2861 case S_IFSOCK:
2862 case 0: /* zero mode translates to S_IFREG */
2863 return 0;
2864 case S_IFDIR:
2865 return -EPERM;
2866 default:
2867 return -EINVAL;
2868 }
2869}
2870
Al Viro8208a222011-07-25 17:32:17 -04002871SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002872 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873{
Al Viro2ad94ae2008-07-21 09:32:51 -04002874 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002875 struct path path;
2876 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
2878 if (S_ISDIR(mode))
2879 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
Al Virodae6ad82011-06-26 11:50:15 -04002881 dentry = user_path_create(dfd, filename, &path, 0);
2882 if (IS_ERR(dentry))
2883 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002884
Al Virodae6ad82011-06-26 11:50:15 -04002885 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002886 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002887 error = may_mknod(mode);
2888 if (error)
2889 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002890 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002891 if (error)
2892 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002893 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002894 if (error)
2895 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002896 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002898 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 break;
2900 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002901 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 new_decode_dev(dev));
2903 break;
2904 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002905 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002908out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002909 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002910out_dput:
2911 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002912 mutex_unlock(&path.dentry->d_inode->i_mutex);
2913 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914
2915 return error;
2916}
2917
Al Viro8208a222011-07-25 17:32:17 -04002918SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002919{
2920 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2921}
2922
Al Viro18bb1db2011-07-26 01:41:39 -04002923int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002925 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05002926 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927
2928 if (error)
2929 return error;
2930
Al Viroacfa4382008-12-04 10:06:33 -05002931 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 return -EPERM;
2933
2934 mode &= (S_IRWXUGO|S_ISVTX);
2935 error = security_inode_mkdir(dir, dentry, mode);
2936 if (error)
2937 return error;
2938
Al Viro8de52772012-02-06 12:45:27 -05002939 if (max_links && dir->i_nlink >= max_links)
2940 return -EMLINK;
2941
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002943 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002944 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 return error;
2946}
2947
Al Viroa218d0f2011-11-21 14:59:34 -05002948SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949{
Dave Hansen6902d922006-09-30 23:29:01 -07002950 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002951 struct path path;
2952 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
Al Virodae6ad82011-06-26 11:50:15 -04002954 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002955 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002956 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002957
Al Virodae6ad82011-06-26 11:50:15 -04002958 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002959 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002960 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002961 if (error)
2962 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002963 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002964 if (error)
2965 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002966 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002967out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002968 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002969out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002970 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002971 mutex_unlock(&path.dentry->d_inode->i_mutex);
2972 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 return error;
2974}
2975
Al Viroa218d0f2011-11-21 14:59:34 -05002976SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002977{
2978 return sys_mkdirat(AT_FDCWD, pathname, mode);
2979}
2980
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981/*
Sage Weila71905f2011-05-24 13:06:08 -07002982 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05002983 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07002984 * dentry, and if that is true (possibly after pruning the dcache),
2985 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 *
2987 * A low-level filesystem can, if it choses, legally
2988 * do a
2989 *
2990 * if (!d_unhashed(dentry))
2991 * return -EBUSY;
2992 *
2993 * if it cannot handle the case of removing a directory
2994 * that is still in use by something else..
2995 */
2996void dentry_unhash(struct dentry *dentry)
2997{
Vasily Averindc168422006-12-06 20:37:07 -08002998 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07003000 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 __d_drop(dentry);
3002 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003}
3004
3005int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3006{
3007 int error = may_delete(dir, dentry, 1);
3008
3009 if (error)
3010 return error;
3011
Al Viroacfa4382008-12-04 10:06:33 -05003012 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 return -EPERM;
3014
Al Viro1d2ef592011-09-14 18:55:41 +01003015 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003016 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Sage Weil912dbc12011-05-24 13:06:11 -07003018 error = -EBUSY;
3019 if (d_mountpoint(dentry))
3020 goto out;
3021
3022 error = security_inode_rmdir(dir, dentry);
3023 if (error)
3024 goto out;
3025
Sage Weil3cebde22011-05-29 21:20:59 -07003026 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003027 error = dir->i_op->rmdir(dir, dentry);
3028 if (error)
3029 goto out;
3030
3031 dentry->d_inode->i_flags |= S_DEAD;
3032 dont_mount(dentry);
3033
3034out:
3035 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003036 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003037 if (!error)
3038 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 return error;
3040}
3041
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003042static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043{
3044 int error = 0;
3045 char * name;
3046 struct dentry *dentry;
3047 struct nameidata nd;
3048
Al Viro2ad94ae2008-07-21 09:32:51 -04003049 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003051 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
3053 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003054 case LAST_DOTDOT:
3055 error = -ENOTEMPTY;
3056 goto exit1;
3057 case LAST_DOT:
3058 error = -EINVAL;
3059 goto exit1;
3060 case LAST_ROOT:
3061 error = -EBUSY;
3062 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003064
3065 nd.flags &= ~LOOKUP_PARENT;
3066
Jan Blunck4ac91372008-02-14 19:34:32 -08003067 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003068 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003070 if (IS_ERR(dentry))
3071 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003072 if (!dentry->d_inode) {
3073 error = -ENOENT;
3074 goto exit3;
3075 }
Dave Hansen06227532008-02-15 14:37:34 -08003076 error = mnt_want_write(nd.path.mnt);
3077 if (error)
3078 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003079 error = security_path_rmdir(&nd.path, dentry);
3080 if (error)
3081 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08003082 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003083exit4:
Dave Hansen06227532008-02-15 14:37:34 -08003084 mnt_drop_write(nd.path.mnt);
3085exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003086 dput(dentry);
3087exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003088 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003090 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 putname(name);
3092 return error;
3093}
3094
Heiko Carstens3cdad422009-01-14 14:14:22 +01003095SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003096{
3097 return do_rmdir(AT_FDCWD, pathname);
3098}
3099
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100int vfs_unlink(struct inode *dir, struct dentry *dentry)
3101{
3102 int error = may_delete(dir, dentry, 0);
3103
3104 if (error)
3105 return error;
3106
Al Viroacfa4382008-12-04 10:06:33 -05003107 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 return -EPERM;
3109
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003110 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 if (d_mountpoint(dentry))
3112 error = -EBUSY;
3113 else {
3114 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003115 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003117 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003118 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003121 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122
3123 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3124 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08003125 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 }
Robert Love0eeca282005-07-12 17:06:03 -04003128
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 return error;
3130}
3131
3132/*
3133 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003134 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 * writeout happening, and we don't want to prevent access to the directory
3136 * while waiting on the I/O.
3137 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003138static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139{
Al Viro2ad94ae2008-07-21 09:32:51 -04003140 int error;
3141 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 struct dentry *dentry;
3143 struct nameidata nd;
3144 struct inode *inode = NULL;
3145
Al Viro2ad94ae2008-07-21 09:32:51 -04003146 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003148 return error;
3149
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 error = -EISDIR;
3151 if (nd.last_type != LAST_NORM)
3152 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003153
3154 nd.flags &= ~LOOKUP_PARENT;
3155
Jan Blunck4ac91372008-02-14 19:34:32 -08003156 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003157 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 error = PTR_ERR(dentry);
3159 if (!IS_ERR(dentry)) {
3160 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003161 if (nd.last.name[nd.last.len])
3162 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03003164 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003165 goto slashes;
3166 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08003167 error = mnt_want_write(nd.path.mnt);
3168 if (error)
3169 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003170 error = security_path_unlink(&nd.path, dentry);
3171 if (error)
3172 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003173 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003174exit3:
Dave Hansen06227532008-02-15 14:37:34 -08003175 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176 exit2:
3177 dput(dentry);
3178 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003179 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180 if (inode)
3181 iput(inode); /* truncate the inode here */
3182exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003183 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 putname(name);
3185 return error;
3186
3187slashes:
3188 error = !dentry->d_inode ? -ENOENT :
3189 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3190 goto exit2;
3191}
3192
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003193SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003194{
3195 if ((flag & ~AT_REMOVEDIR) != 0)
3196 return -EINVAL;
3197
3198 if (flag & AT_REMOVEDIR)
3199 return do_rmdir(dfd, pathname);
3200
3201 return do_unlinkat(dfd, pathname);
3202}
3203
Heiko Carstens3480b252009-01-14 14:14:16 +01003204SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003205{
3206 return do_unlinkat(AT_FDCWD, pathname);
3207}
3208
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003209int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003211 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212
3213 if (error)
3214 return error;
3215
Al Viroacfa4382008-12-04 10:06:33 -05003216 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217 return -EPERM;
3218
3219 error = security_inode_symlink(dir, dentry, oldname);
3220 if (error)
3221 return error;
3222
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003224 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003225 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 return error;
3227}
3228
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003229SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3230 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231{
Al Viro2ad94ae2008-07-21 09:32:51 -04003232 int error;
3233 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003234 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003235 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236
3237 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003238 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003240
Al Virodae6ad82011-06-26 11:50:15 -04003241 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07003242 error = PTR_ERR(dentry);
3243 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003244 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003245
Al Virodae6ad82011-06-26 11:50:15 -04003246 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003247 if (error)
3248 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003249 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003250 if (error)
3251 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003252 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003253out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003254 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003255out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003256 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003257 mutex_unlock(&path.dentry->d_inode->i_mutex);
3258 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07003259out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 putname(from);
3261 return error;
3262}
3263
Heiko Carstens3480b252009-01-14 14:14:16 +01003264SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003265{
3266 return sys_symlinkat(oldname, AT_FDCWD, newname);
3267}
3268
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3270{
3271 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003272 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 int error;
3274
3275 if (!inode)
3276 return -ENOENT;
3277
Miklos Szeredia95164d2008-07-30 15:08:48 +02003278 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 if (error)
3280 return error;
3281
3282 if (dir->i_sb != inode->i_sb)
3283 return -EXDEV;
3284
3285 /*
3286 * A link to an append-only or immutable file cannot be created.
3287 */
3288 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3289 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003290 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003292 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 return -EPERM;
3294
3295 error = security_inode_link(old_dentry, dir, new_dentry);
3296 if (error)
3297 return error;
3298
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003299 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303300 /* Make sure we don't allow creating hardlink to an unlinked file */
3301 if (inode->i_nlink == 0)
3302 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003303 else if (max_links && inode->i_nlink >= max_links)
3304 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303305 else
3306 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003307 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003308 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003309 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 return error;
3311}
3312
3313/*
3314 * Hardlinks are often used in delicate situations. We avoid
3315 * security-related surprises by not following symlinks on the
3316 * newname. --KAB
3317 *
3318 * We don't follow them on the oldname either to be compatible
3319 * with linux 2.0, and to avoid hard-linking to directories
3320 * and other special files. --ADM
3321 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003322SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3323 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324{
3325 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003326 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303327 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303330 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003331 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303332 /*
3333 * To use null names we require CAP_DAC_READ_SEARCH
3334 * This ensures that not everyone will be able to create
3335 * handlink using the passed filedescriptor.
3336 */
3337 if (flags & AT_EMPTY_PATH) {
3338 if (!capable(CAP_DAC_READ_SEARCH))
3339 return -ENOENT;
3340 how = LOOKUP_EMPTY;
3341 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003342
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303343 if (flags & AT_SYMLINK_FOLLOW)
3344 how |= LOOKUP_FOLLOW;
3345
3346 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003348 return error;
3349
Al Virodae6ad82011-06-26 11:50:15 -04003350 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003352 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003353 goto out;
3354
3355 error = -EXDEV;
3356 if (old_path.mnt != new_path.mnt)
3357 goto out_dput;
3358 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003359 if (error)
3360 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003361 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003362 if (error)
3363 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003364 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003365out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003366 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003367out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003368 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003369 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3370 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371out:
Al Viro2d8f3032008-07-22 09:59:21 -04003372 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
3374 return error;
3375}
3376
Heiko Carstens3480b252009-01-14 14:14:16 +01003377SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003378{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003379 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003380}
3381
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382/*
3383 * The worst of all namespace operations - renaming directory. "Perverted"
3384 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3385 * Problems:
3386 * a) we can get into loop creation. Check is done in is_subdir().
3387 * b) race potential - two innocent renames can create a loop together.
3388 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003389 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 * story.
3391 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003392 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 * whether the target exists). Solution: try to be smart with locking
3394 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003395 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 * move will be locked. Thus we can rank directories by the tree
3397 * (ancestors first) and rank all non-directories after them.
3398 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003399 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 * HOWEVER, it relies on the assumption that any object with ->lookup()
3401 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3402 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003403 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003404 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003406 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 * locking].
3408 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003409static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3410 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411{
3412 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003413 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003414 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415
3416 /*
3417 * If we are going to change the parent - check write permissions,
3418 * we'll need to flip '..'.
3419 */
3420 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003421 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 if (error)
3423 return error;
3424 }
3425
3426 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3427 if (error)
3428 return error;
3429
Al Viro1d2ef592011-09-14 18:55:41 +01003430 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003431 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003432 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003433
3434 error = -EBUSY;
3435 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3436 goto out;
3437
Al Viro8de52772012-02-06 12:45:27 -05003438 error = -EMLINK;
3439 if (max_links && !target && new_dir != old_dir &&
3440 new_dir->i_nlink >= max_links)
3441 goto out;
3442
Sage Weil3cebde22011-05-29 21:20:59 -07003443 if (target)
3444 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003445 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3446 if (error)
3447 goto out;
3448
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003450 target->i_flags |= S_DEAD;
3451 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452 }
Sage Weil9055cba2011-05-24 13:06:12 -07003453out:
3454 if (target)
3455 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003456 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003457 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003458 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3459 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 return error;
3461}
3462
Adrian Bunk75c96f82005-05-05 16:16:09 -07003463static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3464 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465{
Sage Weil51892bb2011-05-24 13:06:13 -07003466 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467 int error;
3468
3469 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3470 if (error)
3471 return error;
3472
3473 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003475 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003476
3477 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003479 goto out;
3480
3481 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3482 if (error)
3483 goto out;
3484
3485 if (target)
3486 dont_mount(new_dentry);
3487 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3488 d_move(old_dentry, new_dentry);
3489out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003491 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 dput(new_dentry);
3493 return error;
3494}
3495
3496int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3497 struct inode *new_dir, struct dentry *new_dentry)
3498{
3499 int error;
3500 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003501 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502
3503 if (old_dentry->d_inode == new_dentry->d_inode)
3504 return 0;
3505
3506 error = may_delete(old_dir, old_dentry, is_dir);
3507 if (error)
3508 return error;
3509
3510 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003511 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 else
3513 error = may_delete(new_dir, new_dentry, is_dir);
3514 if (error)
3515 return error;
3516
Al Viroacfa4382008-12-04 10:06:33 -05003517 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518 return -EPERM;
3519
Robert Love0eeca282005-07-12 17:06:03 -04003520 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3521
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 if (is_dir)
3523 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3524 else
3525 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003526 if (!error)
3527 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003528 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003529 fsnotify_oldname_free(old_name);
3530
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 return error;
3532}
3533
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003534SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3535 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536{
Al Viro2ad94ae2008-07-21 09:32:51 -04003537 struct dentry *old_dir, *new_dir;
3538 struct dentry *old_dentry, *new_dentry;
3539 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003541 char *from;
3542 char *to;
3543 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544
Al Viro2ad94ae2008-07-21 09:32:51 -04003545 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546 if (error)
3547 goto exit;
3548
Al Viro2ad94ae2008-07-21 09:32:51 -04003549 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 if (error)
3551 goto exit1;
3552
3553 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003554 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 goto exit2;
3556
Jan Blunck4ac91372008-02-14 19:34:32 -08003557 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558 error = -EBUSY;
3559 if (oldnd.last_type != LAST_NORM)
3560 goto exit2;
3561
Jan Blunck4ac91372008-02-14 19:34:32 -08003562 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 if (newnd.last_type != LAST_NORM)
3564 goto exit2;
3565
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003566 oldnd.flags &= ~LOOKUP_PARENT;
3567 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003568 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003569
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 trap = lock_rename(new_dir, old_dir);
3571
Christoph Hellwig49705b72005-11-08 21:35:06 -08003572 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 error = PTR_ERR(old_dentry);
3574 if (IS_ERR(old_dentry))
3575 goto exit3;
3576 /* source must exist */
3577 error = -ENOENT;
3578 if (!old_dentry->d_inode)
3579 goto exit4;
3580 /* unless the source is a directory trailing slashes give -ENOTDIR */
3581 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3582 error = -ENOTDIR;
3583 if (oldnd.last.name[oldnd.last.len])
3584 goto exit4;
3585 if (newnd.last.name[newnd.last.len])
3586 goto exit4;
3587 }
3588 /* source should not be ancestor of target */
3589 error = -EINVAL;
3590 if (old_dentry == trap)
3591 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003592 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 error = PTR_ERR(new_dentry);
3594 if (IS_ERR(new_dentry))
3595 goto exit4;
3596 /* target should not be an ancestor of source */
3597 error = -ENOTEMPTY;
3598 if (new_dentry == trap)
3599 goto exit5;
3600
Dave Hansen9079b1e2008-02-15 14:37:49 -08003601 error = mnt_want_write(oldnd.path.mnt);
3602 if (error)
3603 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003604 error = security_path_rename(&oldnd.path, old_dentry,
3605 &newnd.path, new_dentry);
3606 if (error)
3607 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 error = vfs_rename(old_dir->d_inode, old_dentry,
3609 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003610exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003611 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612exit5:
3613 dput(new_dentry);
3614exit4:
3615 dput(old_dentry);
3616exit3:
3617 unlock_rename(new_dir, old_dir);
3618exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003619 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003620 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003622 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003624exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 return error;
3626}
3627
Heiko Carstensa26eab22009-01-14 14:14:17 +01003628SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003629{
3630 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3631}
3632
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3634{
3635 int len;
3636
3637 len = PTR_ERR(link);
3638 if (IS_ERR(link))
3639 goto out;
3640
3641 len = strlen(link);
3642 if (len > (unsigned) buflen)
3643 len = buflen;
3644 if (copy_to_user(buffer, link, len))
3645 len = -EFAULT;
3646out:
3647 return len;
3648}
3649
3650/*
3651 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3652 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3653 * using) it for any given inode is up to filesystem.
3654 */
3655int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3656{
3657 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003658 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003659 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003660
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003662 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003663 if (IS_ERR(cookie))
3664 return PTR_ERR(cookie);
3665
3666 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3667 if (dentry->d_inode->i_op->put_link)
3668 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3669 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670}
3671
3672int vfs_follow_link(struct nameidata *nd, const char *link)
3673{
3674 return __vfs_follow_link(nd, link);
3675}
3676
3677/* get the link contents into pagecache */
3678static char *page_getlink(struct dentry * dentry, struct page **ppage)
3679{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003680 char *kaddr;
3681 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003683 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003685 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003687 kaddr = kmap(page);
3688 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3689 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690}
3691
3692int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3693{
3694 struct page *page = NULL;
3695 char *s = page_getlink(dentry, &page);
3696 int res = vfs_readlink(dentry,buffer,buflen,s);
3697 if (page) {
3698 kunmap(page);
3699 page_cache_release(page);
3700 }
3701 return res;
3702}
3703
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003704void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003706 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003708 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709}
3710
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003711void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003713 struct page *page = cookie;
3714
3715 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 kunmap(page);
3717 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 }
3719}
3720
Nick Piggin54566b22009-01-04 12:00:53 -08003721/*
3722 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3723 */
3724int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725{
3726 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003727 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003728 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003729 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003731 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3732 if (nofs)
3733 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734
NeilBrown7e53cac2006-03-25 03:07:57 -08003735retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003736 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003737 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003739 goto fail;
3740
Cong Wange8e3c3d2011-11-25 23:14:27 +08003741 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08003743 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07003744
3745 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3746 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 if (err < 0)
3748 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003749 if (err < len-1)
3750 goto retry;
3751
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752 mark_inode_dirty(inode);
3753 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754fail:
3755 return err;
3756}
3757
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003758int page_symlink(struct inode *inode, const char *symname, int len)
3759{
3760 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003761 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003762}
3763
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003764const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 .readlink = generic_readlink,
3766 .follow_link = page_follow_link_light,
3767 .put_link = page_put_link,
3768};
3769
Al Viro2d8f3032008-07-22 09:59:21 -04003770EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003771EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772EXPORT_SYMBOL(follow_down);
3773EXPORT_SYMBOL(follow_up);
3774EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3775EXPORT_SYMBOL(getname);
3776EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777EXPORT_SYMBOL(lookup_one_len);
3778EXPORT_SYMBOL(page_follow_link_light);
3779EXPORT_SYMBOL(page_put_link);
3780EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003781EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782EXPORT_SYMBOL(page_symlink);
3783EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003784EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003785EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003786EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787EXPORT_SYMBOL(unlock_rename);
3788EXPORT_SYMBOL(vfs_create);
3789EXPORT_SYMBOL(vfs_follow_link);
3790EXPORT_SYMBOL(vfs_link);
3791EXPORT_SYMBOL(vfs_mkdir);
3792EXPORT_SYMBOL(vfs_mknod);
3793EXPORT_SYMBOL(generic_permission);
3794EXPORT_SYMBOL(vfs_readlink);
3795EXPORT_SYMBOL(vfs_rename);
3796EXPORT_SYMBOL(vfs_rmdir);
3797EXPORT_SYMBOL(vfs_symlink);
3798EXPORT_SYMBOL(vfs_unlink);
3799EXPORT_SYMBOL(dentry_unhash);
3800EXPORT_SYMBOL(generic_readlink);