blob: 7d694194024ac4d2459e7cc3d60014bdff64e3ba [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
Nick Piggin31e6b012011-01-07 17:49:52 +1100466/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700467 * release_open_intent - free up open intent resources
468 * @nd: pointer to nameidata
469 */
470void release_open_intent(struct nameidata *nd)
471{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800472 struct file *file = nd->intent.open.file;
473
474 if (file && !IS_ERR(file)) {
475 if (file->f_path.dentry == NULL)
476 put_filp(file);
477 else
478 fput(file);
479 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700480}
481
Al Virof60aef72011-02-15 01:35:28 -0500482static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100483{
Al Virof60aef72011-02-15 01:35:28 -0500484 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100485}
486
Al Viro9f1fafe2011-03-25 11:00:12 -0400487/**
488 * complete_walk - successful completion of path walk
489 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500490 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400491 * If we had been in RCU mode, drop out of it and legitimize nd->path.
492 * Revalidate the final result, unless we'd already done that during
493 * the path walk or the filesystem doesn't ask for it. Return 0 on
494 * success, -error on failure. In case of failure caller does not
495 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500496 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400497static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500498{
Al Viro16c2cd72011-02-22 15:50:10 -0500499 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500500 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500501
Al Viro9f1fafe2011-03-25 11:00:12 -0400502 if (nd->flags & LOOKUP_RCU) {
503 nd->flags &= ~LOOKUP_RCU;
504 if (!(nd->flags & LOOKUP_ROOT))
505 nd->root.mnt = NULL;
506 spin_lock(&dentry->d_lock);
507 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
508 spin_unlock(&dentry->d_lock);
509 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930510 br_read_unlock(&vfsmount_lock);
Al Viro9f1fafe2011-03-25 11:00:12 -0400511 return -ECHILD;
512 }
513 BUG_ON(nd->inode != dentry->d_inode);
514 spin_unlock(&dentry->d_lock);
515 mntget(nd->path.mnt);
516 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930517 br_read_unlock(&vfsmount_lock);
Al Viro9f1fafe2011-03-25 11:00:12 -0400518 }
519
Al Viro16c2cd72011-02-22 15:50:10 -0500520 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500521 return 0;
522
Al Viro16c2cd72011-02-22 15:50:10 -0500523 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
524 return 0;
525
526 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
527 return 0;
528
529 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100530 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500531 if (status > 0)
532 return 0;
533
Al Viro16c2cd72011-02-22 15:50:10 -0500534 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500535 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500536
Al Viro9f1fafe2011-03-25 11:00:12 -0400537 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500538 return status;
539}
540
Al Viro2a737872009-04-07 11:49:53 -0400541static __always_inline void set_root(struct nameidata *nd)
542{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200543 if (!nd->root.mnt)
544 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400545}
546
Al Viro6de88d72009-08-09 01:41:57 +0400547static int link_path_walk(const char *, struct nameidata *);
548
Nick Piggin31e6b012011-01-07 17:49:52 +1100549static __always_inline void set_root_rcu(struct nameidata *nd)
550{
551 if (!nd->root.mnt) {
552 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100553 unsigned seq;
554
555 do {
556 seq = read_seqcount_begin(&fs->seq);
557 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700558 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100559 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100560 }
561}
562
Arjan van de Venf1662352006-01-14 13:21:31 -0800563static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
Nick Piggin31e6b012011-01-07 17:49:52 +1100565 int ret;
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 if (IS_ERR(link))
568 goto fail;
569
570 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400571 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800572 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400573 nd->path = nd->root;
574 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500575 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100577 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100578
Nick Piggin31e6b012011-01-07 17:49:52 +1100579 ret = link_path_walk(link, nd);
580 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800582 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return PTR_ERR(link);
584}
585
Jan Blunck1d957f92008-02-14 19:34:35 -0800586static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700587{
588 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800589 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700590 mntput(path->mnt);
591}
592
Nick Piggin7b9337a2011-01-14 08:42:43 +0000593static inline void path_to_nameidata(const struct path *path,
594 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700595{
Nick Piggin31e6b012011-01-07 17:49:52 +1100596 if (!(nd->flags & LOOKUP_RCU)) {
597 dput(nd->path.dentry);
598 if (nd->path.mnt != path->mnt)
599 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800600 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100601 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800602 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700603}
604
Al Viro574197e2011-03-14 22:20:34 -0400605static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
606{
607 struct inode *inode = link->dentry->d_inode;
608 if (!IS_ERR(cookie) && inode->i_op->put_link)
609 inode->i_op->put_link(link->dentry, nd, cookie);
610 path_put(link);
611}
612
Al Virodef4af32009-12-26 08:37:05 -0500613static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400614follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800615{
616 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000617 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800618
Al Viro844a3912011-02-15 00:38:26 -0500619 BUG_ON(nd->flags & LOOKUP_RCU);
620
Al Viro0e794582011-03-16 02:45:02 -0400621 if (link->mnt == nd->path.mnt)
622 mntget(link->mnt);
623
Al Viro574197e2011-03-14 22:20:34 -0400624 if (unlikely(current->total_link_count >= 40)) {
625 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400626 path_put(&nd->path);
627 return -ELOOP;
628 }
629 cond_resched();
630 current->total_link_count++;
631
Al Viro68ac1232012-03-15 08:21:57 -0400632 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800633 nd_set_link(nd, NULL);
634
Al Viro36f3b4f2011-02-22 21:24:38 -0500635 error = security_inode_follow_link(link->dentry, nd);
636 if (error) {
637 *p = ERR_PTR(error); /* no ->put_link(), please */
638 path_put(&nd->path);
639 return error;
640 }
641
Al Viro86acdca12009-12-22 23:45:11 -0500642 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500643 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
644 error = PTR_ERR(*p);
645 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800646 char *s = nd_get_link(nd);
647 error = 0;
648 if (s)
649 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400650 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500651 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400652 nd->inode = nd->path.dentry->d_inode;
653 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400654 /* stepped on a _really_ weird one */
655 path_put(&nd->path);
656 error = -ELOOP;
657 }
658 }
Ian Kent051d3812006-03-27 01:14:53 -0800659 }
Ian Kent051d3812006-03-27 01:14:53 -0800660 return error;
661}
662
Nick Piggin31e6b012011-01-07 17:49:52 +1100663static int follow_up_rcu(struct path *path)
664{
Al Viro0714a532011-11-24 22:19:58 -0500665 struct mount *mnt = real_mount(path->mnt);
666 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100667 struct dentry *mountpoint;
668
Al Viro0714a532011-11-24 22:19:58 -0500669 parent = mnt->mnt_parent;
670 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100671 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500672 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100673 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500674 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100675 return 1;
676}
677
Al Virobab77eb2009-04-18 03:26:48 -0400678int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Al Viro0714a532011-11-24 22:19:58 -0500680 struct mount *mnt = real_mount(path->mnt);
681 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000683
Andi Kleen962830d2012-05-08 13:32:02 +0930684 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500685 parent = mnt->mnt_parent;
686 if (&parent->mnt == path->mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930687 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return 0;
689 }
Al Viro0714a532011-11-24 22:19:58 -0500690 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500691 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930692 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400693 dput(path->dentry);
694 path->dentry = mountpoint;
695 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500696 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return 1;
698}
699
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100700/*
David Howells9875cf82011-01-14 18:45:21 +0000701 * Perform an automount
702 * - return -EISDIR to tell follow_managed() to stop and return the path we
703 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 */
David Howells9875cf82011-01-14 18:45:21 +0000705static int follow_automount(struct path *path, unsigned flags,
706 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100707{
David Howells9875cf82011-01-14 18:45:21 +0000708 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000709 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100710
David Howells9875cf82011-01-14 18:45:21 +0000711 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
712 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700713
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200714 /* We don't want to mount if someone's just doing a stat -
715 * unless they're stat'ing a directory and appended a '/' to
716 * the name.
717 *
718 * We do, however, want to mount if someone wants to open or
719 * create a file of any type under the mountpoint, wants to
720 * traverse through the mountpoint or wants to open the
721 * mounted directory. Also, autofs may mark negative dentries
722 * as being automount points. These will need the attentions
723 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000724 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200725 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700726 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200727 path->dentry->d_inode)
728 return -EISDIR;
729
David Howells9875cf82011-01-14 18:45:21 +0000730 current->total_link_count++;
731 if (current->total_link_count >= 40)
732 return -ELOOP;
733
734 mnt = path->dentry->d_op->d_automount(path);
735 if (IS_ERR(mnt)) {
736 /*
737 * The filesystem is allowed to return -EISDIR here to indicate
738 * it doesn't want to automount. For instance, autofs would do
739 * this so that its userspace daemon can mount on this dentry.
740 *
741 * However, we can only permit this if it's a terminal point in
742 * the path being looked up; if it wasn't then the remainder of
743 * the path is inaccessible and we should say so.
744 */
Al Viro49084c32011-06-25 21:59:52 -0400745 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000746 return -EREMOTE;
747 return PTR_ERR(mnt);
748 }
David Howellsea5b7782011-01-14 19:10:03 +0000749
David Howells9875cf82011-01-14 18:45:21 +0000750 if (!mnt) /* mount collision */
751 return 0;
752
Al Viro8aef1882011-06-16 15:10:06 +0100753 if (!*need_mntput) {
754 /* lock_mount() may release path->mnt on error */
755 mntget(path->mnt);
756 *need_mntput = true;
757 }
Al Viro19a167a2011-01-17 01:35:23 -0500758 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000759
David Howellsea5b7782011-01-14 19:10:03 +0000760 switch (err) {
761 case -EBUSY:
762 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500763 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000764 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100765 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000766 path->mnt = mnt;
767 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000768 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500769 default:
770 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000771 }
Al Viro19a167a2011-01-17 01:35:23 -0500772
David Howells9875cf82011-01-14 18:45:21 +0000773}
774
775/*
776 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000777 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000778 * - Flagged as mountpoint
779 * - Flagged as automount point
780 *
781 * This may only be called in refwalk mode.
782 *
783 * Serialization is taken care of in namespace.c
784 */
785static int follow_managed(struct path *path, unsigned flags)
786{
Al Viro8aef1882011-06-16 15:10:06 +0100787 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000788 unsigned managed;
789 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100790 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000791
792 /* Given that we're not holding a lock here, we retain the value in a
793 * local variable for each dentry as we look at it so that we don't see
794 * the components of that value change under us */
795 while (managed = ACCESS_ONCE(path->dentry->d_flags),
796 managed &= DCACHE_MANAGED_DENTRY,
797 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000798 /* Allow the filesystem to manage the transit without i_mutex
799 * being held. */
800 if (managed & DCACHE_MANAGE_TRANSIT) {
801 BUG_ON(!path->dentry->d_op);
802 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400803 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000804 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100805 break;
David Howellscc53ce52011-01-14 18:45:26 +0000806 }
807
David Howells9875cf82011-01-14 18:45:21 +0000808 /* Transit to a mounted filesystem. */
809 if (managed & DCACHE_MOUNTED) {
810 struct vfsmount *mounted = lookup_mnt(path);
811 if (mounted) {
812 dput(path->dentry);
813 if (need_mntput)
814 mntput(path->mnt);
815 path->mnt = mounted;
816 path->dentry = dget(mounted->mnt_root);
817 need_mntput = true;
818 continue;
819 }
820
821 /* Something is mounted on this dentry in another
822 * namespace and/or whatever was mounted there in this
823 * namespace got unmounted before we managed to get the
824 * vfsmount_lock */
825 }
826
827 /* Handle an automount point */
828 if (managed & DCACHE_NEED_AUTOMOUNT) {
829 ret = follow_automount(path, flags, &need_mntput);
830 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100831 break;
David Howells9875cf82011-01-14 18:45:21 +0000832 continue;
833 }
834
835 /* We didn't change the current path point */
836 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
Al Viro8aef1882011-06-16 15:10:06 +0100838
839 if (need_mntput && path->mnt == mnt)
840 mntput(path->mnt);
841 if (ret == -EISDIR)
842 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +0000843 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844}
845
David Howellscc53ce52011-01-14 18:45:26 +0000846int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847{
848 struct vfsmount *mounted;
849
Al Viro1c755af2009-04-18 14:06:57 -0400850 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400852 dput(path->dentry);
853 mntput(path->mnt);
854 path->mnt = mounted;
855 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 return 1;
857 }
858 return 0;
859}
860
Ian Kent62a73752011-03-25 01:51:02 +0800861static inline bool managed_dentry_might_block(struct dentry *dentry)
862{
863 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
864 dentry->d_op->d_manage(dentry, true) < 0);
865}
866
David Howells9875cf82011-01-14 18:45:21 +0000867/*
Al Viro287548e2011-05-27 06:50:06 -0400868 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
869 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000870 */
871static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400872 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000873{
Ian Kent62a73752011-03-25 01:51:02 +0800874 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -0500875 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800876 /*
877 * Don't forget we might have a non-mountpoint managed dentry
878 * that wants to block transit.
879 */
Al Viro287548e2011-05-27 06:50:06 -0400880 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000881 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800882
883 if (!d_mountpoint(path->dentry))
884 break;
885
David Howells9875cf82011-01-14 18:45:21 +0000886 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
887 if (!mounted)
888 break;
Al Viroc7105362011-11-24 18:22:03 -0500889 path->mnt = &mounted->mnt;
890 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +0000891 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +0000892 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700893 /*
894 * Update the inode too. We don't need to re-check the
895 * dentry sequence number here after this d_inode read,
896 * because a mount-point is always pinned.
897 */
898 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000899 }
David Howells9875cf82011-01-14 18:45:21 +0000900 return true;
901}
902
Al Virodea39372011-05-27 06:53:39 -0400903static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400904{
Al Virodea39372011-05-27 06:53:39 -0400905 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -0500906 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400907 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400908 if (!mounted)
909 break;
Al Viroc7105362011-11-24 18:22:03 -0500910 nd->path.mnt = &mounted->mnt;
911 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -0400912 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400913 }
914}
915
Nick Piggin31e6b012011-01-07 17:49:52 +1100916static int follow_dotdot_rcu(struct nameidata *nd)
917{
Nick Piggin31e6b012011-01-07 17:49:52 +1100918 set_root_rcu(nd);
919
David Howells9875cf82011-01-14 18:45:21 +0000920 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100921 if (nd->path.dentry == nd->root.dentry &&
922 nd->path.mnt == nd->root.mnt) {
923 break;
924 }
925 if (nd->path.dentry != nd->path.mnt->mnt_root) {
926 struct dentry *old = nd->path.dentry;
927 struct dentry *parent = old->d_parent;
928 unsigned seq;
929
930 seq = read_seqcount_begin(&parent->d_seq);
931 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500932 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100933 nd->path.dentry = parent;
934 nd->seq = seq;
935 break;
936 }
937 if (!follow_up_rcu(&nd->path))
938 break;
939 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100940 }
Al Virodea39372011-05-27 06:53:39 -0400941 follow_mount_rcu(nd);
942 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100943 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500944
945failed:
946 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500947 if (!(nd->flags & LOOKUP_ROOT))
948 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500949 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930950 br_read_unlock(&vfsmount_lock);
Al Viroef7562d2011-03-04 14:35:59 -0500951 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100952}
953
David Howells9875cf82011-01-14 18:45:21 +0000954/*
David Howellscc53ce52011-01-14 18:45:26 +0000955 * Follow down to the covering mount currently visible to userspace. At each
956 * point, the filesystem owning that dentry may be queried as to whether the
957 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000958 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400959int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000960{
961 unsigned managed;
962 int ret;
963
964 while (managed = ACCESS_ONCE(path->dentry->d_flags),
965 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
966 /* Allow the filesystem to manage the transit without i_mutex
967 * being held.
968 *
969 * We indicate to the filesystem if someone is trying to mount
970 * something here. This gives autofs the chance to deny anyone
971 * other than its daemon the right to mount on its
972 * superstructure.
973 *
974 * The filesystem may sleep at this point.
975 */
976 if (managed & DCACHE_MANAGE_TRANSIT) {
977 BUG_ON(!path->dentry->d_op);
978 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000979 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400980 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000981 if (ret < 0)
982 return ret == -EISDIR ? 0 : ret;
983 }
984
985 /* Transit to a mounted filesystem. */
986 if (managed & DCACHE_MOUNTED) {
987 struct vfsmount *mounted = lookup_mnt(path);
988 if (!mounted)
989 break;
990 dput(path->dentry);
991 mntput(path->mnt);
992 path->mnt = mounted;
993 path->dentry = dget(mounted->mnt_root);
994 continue;
995 }
996
997 /* Don't handle automount points here */
998 break;
999 }
1000 return 0;
1001}
1002
1003/*
David Howells9875cf82011-01-14 18:45:21 +00001004 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1005 */
1006static void follow_mount(struct path *path)
1007{
1008 while (d_mountpoint(path->dentry)) {
1009 struct vfsmount *mounted = lookup_mnt(path);
1010 if (!mounted)
1011 break;
1012 dput(path->dentry);
1013 mntput(path->mnt);
1014 path->mnt = mounted;
1015 path->dentry = dget(mounted->mnt_root);
1016 }
1017}
1018
Nick Piggin31e6b012011-01-07 17:49:52 +11001019static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
Al Viro2a737872009-04-07 11:49:53 -04001021 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001024 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Al Viro2a737872009-04-07 11:49:53 -04001026 if (nd->path.dentry == nd->root.dentry &&
1027 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 break;
1029 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001030 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001031 /* rare case of legitimate dget_parent()... */
1032 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 dput(old);
1034 break;
1035 }
Al Viro3088dd72010-01-30 15:47:29 -05001036 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
Al Viro79ed0222009-04-18 13:59:41 -04001039 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001040 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041}
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001044 * This looks up the name in dcache, possibly revalidates the old dentry and
1045 * allocates a new one if not found or not valid. In the need_lookup argument
1046 * returns whether i_op->lookup is necessary.
1047 *
1048 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001049 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001050static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1051 struct nameidata *nd, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001052{
Nick Pigginbaa03892010-08-18 04:37:31 +10001053 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001054 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001055
Miklos Szeredibad61182012-03-26 12:54:24 +02001056 *need_lookup = false;
1057 dentry = d_lookup(dir, name);
1058 if (dentry) {
1059 if (d_need_lookup(dentry)) {
1060 *need_lookup = true;
1061 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1062 error = d_revalidate(dentry, nd);
1063 if (unlikely(error <= 0)) {
1064 if (error < 0) {
1065 dput(dentry);
1066 return ERR_PTR(error);
1067 } else if (!d_invalidate(dentry)) {
1068 dput(dentry);
1069 dentry = NULL;
1070 }
1071 }
1072 }
1073 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001074
Miklos Szeredibad61182012-03-26 12:54:24 +02001075 if (!dentry) {
1076 dentry = d_alloc(dir, name);
1077 if (unlikely(!dentry))
1078 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001079
Miklos Szeredibad61182012-03-26 12:54:24 +02001080 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001081 }
1082 return dentry;
1083}
1084
1085/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001086 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1087 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1088 *
1089 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001090 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001091static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1092 struct nameidata *nd)
Josef Bacik44396f42011-05-31 11:58:49 -04001093{
Josef Bacik44396f42011-05-31 11:58:49 -04001094 struct dentry *old;
1095
1096 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001097 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001098 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001099 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001100 }
Josef Bacik44396f42011-05-31 11:58:49 -04001101
Miklos Szeredibad61182012-03-26 12:54:24 +02001102 old = dir->i_op->lookup(dir, dentry, nd);
Josef Bacik44396f42011-05-31 11:58:49 -04001103 if (unlikely(old)) {
1104 dput(dentry);
1105 dentry = old;
1106 }
1107 return dentry;
1108}
1109
Al Viroa3255542012-03-30 14:41:51 -04001110static struct dentry *__lookup_hash(struct qstr *name,
1111 struct dentry *base, struct nameidata *nd)
1112{
Miklos Szeredibad61182012-03-26 12:54:24 +02001113 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001114 struct dentry *dentry;
1115
Miklos Szeredibad61182012-03-26 12:54:24 +02001116 dentry = lookup_dcache(name, base, nd, &need_lookup);
1117 if (!need_lookup)
1118 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001119
Miklos Szeredibad61182012-03-26 12:54:24 +02001120 return lookup_real(base->d_inode, dentry, nd);
Al Viroa3255542012-03-30 14:41:51 -04001121}
1122
Josef Bacik44396f42011-05-31 11:58:49 -04001123/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 * It's more convoluted than I'd like it to be, but... it's still fairly
1125 * small and for now I'd prefer to have fast path as straight as possible.
1126 * It _is_ time-critical.
1127 */
Miklos Szeredi697f5142012-05-21 17:30:05 +02001128static int lookup_fast(struct nameidata *nd, struct qstr *name,
1129 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130{
Jan Blunck4ac91372008-02-14 19:34:32 -08001131 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001132 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001133 int need_reval = 1;
1134 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001135 int err;
1136
Al Viro3cac2602009-08-13 18:27:43 +04001137 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001138 * Rename seqlock is not required here because in the off chance
1139 * of a false negative due to a concurrent rename, we're going to
1140 * do the non-racy lookup, below.
1141 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001142 if (nd->flags & LOOKUP_RCU) {
1143 unsigned seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001144 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001145 if (!dentry)
1146 goto unlazy;
1147
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001148 /*
1149 * This sequence count validates that the inode matches
1150 * the dentry name information from lookup.
1151 */
1152 *inode = dentry->d_inode;
1153 if (read_seqcount_retry(&dentry->d_seq, seq))
1154 return -ECHILD;
1155
1156 /*
1157 * This sequence count validates that the parent had no
1158 * changes while we did the lookup of the dentry above.
1159 *
1160 * The memory barrier in read_seqcount_begin of child is
1161 * enough, we can use __read_seqcount_retry here.
1162 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001165 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001166
Miklos Szeredifa4ee152012-03-26 12:54:19 +02001167 if (unlikely(d_need_lookup(dentry)))
1168 goto unlazy;
Al Viro24643082011-02-15 01:26:22 -05001169 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001170 status = d_revalidate(dentry, nd);
1171 if (unlikely(status <= 0)) {
1172 if (status != -ECHILD)
1173 need_reval = 0;
1174 goto unlazy;
1175 }
Al Viro24643082011-02-15 01:26:22 -05001176 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001177 path->mnt = mnt;
1178 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001179 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1180 goto unlazy;
1181 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1182 goto unlazy;
1183 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001184unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001185 if (unlazy_walk(nd, dentry))
1186 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001187 } else {
1188 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001189 }
Al Viro5a18fff2011-03-11 04:44:53 -05001190
Al Viro81e6f522012-03-30 14:48:04 -04001191 if (unlikely(!dentry))
1192 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001193
Al Viro81e6f522012-03-30 14:48:04 -04001194 if (unlikely(d_need_lookup(dentry))) {
1195 dput(dentry);
1196 goto need_lookup;
Al Viro24643082011-02-15 01:26:22 -05001197 }
Al Viro81e6f522012-03-30 14:48:04 -04001198
Al Viro5a18fff2011-03-11 04:44:53 -05001199 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1200 status = d_revalidate(dentry, nd);
1201 if (unlikely(status <= 0)) {
1202 if (status < 0) {
1203 dput(dentry);
1204 return status;
1205 }
1206 if (!d_invalidate(dentry)) {
1207 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001208 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001209 }
1210 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001211
David Howells9875cf82011-01-14 18:45:21 +00001212 path->mnt = mnt;
1213 path->dentry = dentry;
1214 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001215 if (unlikely(err < 0)) {
1216 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001217 return err;
Ian Kent89312212011-01-18 12:06:10 +08001218 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001219 if (err)
1220 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001221 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001223
1224need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001225 return 1;
1226}
1227
1228/* Fast lookup failed, do it the slow way */
1229static int lookup_slow(struct nameidata *nd, struct qstr *name,
1230 struct path *path)
1231{
1232 struct dentry *dentry, *parent;
1233 int err;
1234
1235 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001236 BUG_ON(nd->inode != parent->d_inode);
1237
1238 mutex_lock(&parent->d_inode->i_mutex);
1239 dentry = __lookup_hash(name, parent, nd);
1240 mutex_unlock(&parent->d_inode->i_mutex);
1241 if (IS_ERR(dentry))
1242 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001243 path->mnt = nd->path.mnt;
1244 path->dentry = dentry;
1245 err = follow_managed(path, nd->flags);
1246 if (unlikely(err < 0)) {
1247 path_put_conditional(path, nd);
1248 return err;
1249 }
1250 if (err)
1251 nd->flags |= LOOKUP_JUMPED;
1252 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
Al Viro52094c82011-02-21 21:34:47 -05001255static inline int may_lookup(struct nameidata *nd)
1256{
1257 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001258 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001259 if (err != -ECHILD)
1260 return err;
Al Viro19660af2011-03-25 10:32:48 -04001261 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001262 return -ECHILD;
1263 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001264 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001265}
1266
Al Viro9856fa12011-03-04 14:22:06 -05001267static inline int handle_dots(struct nameidata *nd, int type)
1268{
1269 if (type == LAST_DOTDOT) {
1270 if (nd->flags & LOOKUP_RCU) {
1271 if (follow_dotdot_rcu(nd))
1272 return -ECHILD;
1273 } else
1274 follow_dotdot(nd);
1275 }
1276 return 0;
1277}
1278
Al Viro951361f2011-03-04 14:44:37 -05001279static void terminate_walk(struct nameidata *nd)
1280{
1281 if (!(nd->flags & LOOKUP_RCU)) {
1282 path_put(&nd->path);
1283 } else {
1284 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001285 if (!(nd->flags & LOOKUP_ROOT))
1286 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001287 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301288 br_read_unlock(&vfsmount_lock);
Al Viro951361f2011-03-04 14:44:37 -05001289 }
1290}
1291
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001292/*
1293 * Do we need to follow links? We _really_ want to be able
1294 * to do this check without having to look at inode->i_op,
1295 * so we keep a cache of "no, this doesn't need follow_link"
1296 * for the common case.
1297 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001298static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001299{
1300 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1301 if (likely(inode->i_op->follow_link))
1302 return follow;
1303
1304 /* This gets set once for the inode lifetime */
1305 spin_lock(&inode->i_lock);
1306 inode->i_opflags |= IOP_NOFOLLOW;
1307 spin_unlock(&inode->i_lock);
1308 }
1309 return 0;
1310}
1311
Al Viroce57dfc2011-03-13 19:58:58 -04001312static inline int walk_component(struct nameidata *nd, struct path *path,
1313 struct qstr *name, int type, int follow)
1314{
1315 struct inode *inode;
1316 int err;
1317 /*
1318 * "." and ".." are special - ".." especially so because it has
1319 * to be able to know about the current root directory and
1320 * parent relationships.
1321 */
1322 if (unlikely(type != LAST_NORM))
1323 return handle_dots(nd, type);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001324 err = lookup_fast(nd, name, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001325 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001326 if (err < 0)
1327 goto out_err;
1328
1329 err = lookup_slow(nd, name, path);
1330 if (err < 0)
1331 goto out_err;
1332
1333 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001334 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001335 err = -ENOENT;
1336 if (!inode)
1337 goto out_path_put;
1338
Linus Torvalds7813b942011-08-07 09:53:20 -07001339 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001340 if (nd->flags & LOOKUP_RCU) {
1341 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001342 err = -ECHILD;
1343 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001344 }
1345 }
Al Viroce57dfc2011-03-13 19:58:58 -04001346 BUG_ON(inode != path->dentry->d_inode);
1347 return 1;
1348 }
1349 path_to_nameidata(path, nd);
1350 nd->inode = inode;
1351 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001352
1353out_path_put:
1354 path_to_nameidata(path, nd);
1355out_err:
1356 terminate_walk(nd);
1357 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001358}
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360/*
Al Virob3563792011-03-14 21:54:55 -04001361 * This limits recursive symlink follows to 8, while
1362 * limiting consecutive symlinks to 40.
1363 *
1364 * Without that kind of total limit, nasty chains of consecutive
1365 * symlinks can cause almost arbitrarily long lookups.
1366 */
1367static inline int nested_symlink(struct path *path, struct nameidata *nd)
1368{
1369 int res;
1370
Al Virob3563792011-03-14 21:54:55 -04001371 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1372 path_put_conditional(path, nd);
1373 path_put(&nd->path);
1374 return -ELOOP;
1375 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001376 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001377
1378 nd->depth++;
1379 current->link_count++;
1380
1381 do {
1382 struct path link = *path;
1383 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001384
1385 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001386 if (!res)
1387 res = walk_component(nd, path, &nd->last,
1388 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001389 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001390 } while (res > 0);
1391
1392 current->link_count--;
1393 nd->depth--;
1394 return res;
1395}
1396
1397/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001398 * We really don't want to look at inode->i_op->lookup
1399 * when we don't have to. So we keep a cache bit in
1400 * the inode ->i_opflags field that says "yes, we can
1401 * do lookup on this inode".
1402 */
1403static inline int can_lookup(struct inode *inode)
1404{
1405 if (likely(inode->i_opflags & IOP_LOOKUP))
1406 return 1;
1407 if (likely(!inode->i_op->lookup))
1408 return 0;
1409
1410 /* We do this once for the lifetime of the inode */
1411 spin_lock(&inode->i_lock);
1412 inode->i_opflags |= IOP_LOOKUP;
1413 spin_unlock(&inode->i_lock);
1414 return 1;
1415}
1416
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001417/*
1418 * We can do the critical dentry name comparison and hashing
1419 * operations one word at a time, but we are limited to:
1420 *
1421 * - Architectures with fast unaligned word accesses. We could
1422 * do a "get_unaligned()" if this helps and is sufficiently
1423 * fast.
1424 *
1425 * - Little-endian machines (so that we can generate the mask
1426 * of low bytes efficiently). Again, we *could* do a byte
1427 * swapping load on big-endian architectures if that is not
1428 * expensive enough to make the optimization worthless.
1429 *
1430 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1431 * do not trap on the (extremely unlikely) case of a page
1432 * crossing operation.
1433 *
1434 * - Furthermore, we need an efficient 64-bit compile for the
1435 * 64-bit case in order to generate the "number of bytes in
1436 * the final mask". Again, that could be replaced with a
1437 * efficient population count instruction or similar.
1438 */
1439#ifdef CONFIG_DCACHE_WORD_ACCESS
1440
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001441#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001442
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001443#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001444
1445static inline unsigned int fold_hash(unsigned long hash)
1446{
1447 hash += hash >> (8*sizeof(int));
1448 return hash;
1449}
1450
1451#else /* 32-bit case */
1452
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001453#define fold_hash(x) (x)
1454
1455#endif
1456
1457unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1458{
1459 unsigned long a, mask;
1460 unsigned long hash = 0;
1461
1462 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001463 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001464 if (len < sizeof(unsigned long))
1465 break;
1466 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001467 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001468 name += sizeof(unsigned long);
1469 len -= sizeof(unsigned long);
1470 if (!len)
1471 goto done;
1472 }
1473 mask = ~(~0ul << len*8);
1474 hash += mask & a;
1475done:
1476 return fold_hash(hash);
1477}
1478EXPORT_SYMBOL(full_name_hash);
1479
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001480/*
1481 * Calculate the length and hash of the path component, and
1482 * return the length of the component;
1483 */
1484static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1485{
Linus Torvalds36126f82012-05-26 10:43:17 -07001486 unsigned long a, b, adata, bdata, mask, hash, len;
1487 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001488
1489 hash = a = 0;
1490 len = -sizeof(unsigned long);
1491 do {
1492 hash = (hash + a) * 9;
1493 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001494 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001495 b = a ^ REPEAT_BYTE('/');
1496 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001497
Linus Torvalds36126f82012-05-26 10:43:17 -07001498 adata = prep_zero_mask(a, adata, &constants);
1499 bdata = prep_zero_mask(b, bdata, &constants);
1500
1501 mask = create_zero_mask(adata | bdata);
1502
1503 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001504 *hashp = fold_hash(hash);
1505
Linus Torvalds36126f82012-05-26 10:43:17 -07001506 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001507}
1508
1509#else
1510
Linus Torvalds0145acc2012-03-02 14:32:59 -08001511unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1512{
1513 unsigned long hash = init_name_hash();
1514 while (len--)
1515 hash = partial_name_hash(*name++, hash);
1516 return end_name_hash(hash);
1517}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001518EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001519
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001520/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001521 * We know there's a real path component here of at least
1522 * one character.
1523 */
1524static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1525{
1526 unsigned long hash = init_name_hash();
1527 unsigned long len = 0, c;
1528
1529 c = (unsigned char)*name;
1530 do {
1531 len++;
1532 hash = partial_name_hash(c, hash);
1533 c = (unsigned char)name[len];
1534 } while (c && c != '/');
1535 *hashp = end_name_hash(hash);
1536 return len;
1537}
1538
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001539#endif
1540
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001541/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001543 * This is the basic name resolution function, turning a pathname into
1544 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001546 * Returns 0 and nd will have valid dentry and mnt on success.
1547 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 */
Al Viro6de88d72009-08-09 01:41:57 +04001549static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 while (*name=='/')
1555 name++;
1556 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001557 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 /* At this point we know we have a real path component. */
1560 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001562 long len;
Al Virofe479a52011-02-22 15:10:03 -05001563 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Al Viro52094c82011-02-21 21:34:47 -05001565 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (err)
1567 break;
1568
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001569 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001571 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Al Virofe479a52011-02-22 15:10:03 -05001573 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001574 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001575 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001576 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001577 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001578 nd->flags |= LOOKUP_JUMPED;
1579 }
Al Virofe479a52011-02-22 15:10:03 -05001580 break;
1581 case 1:
1582 type = LAST_DOT;
1583 }
Al Viro5a202bc2011-03-08 14:17:44 -05001584 if (likely(type == LAST_NORM)) {
1585 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001586 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001587 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1588 err = parent->d_op->d_hash(parent, nd->inode,
1589 &this);
1590 if (err < 0)
1591 break;
1592 }
1593 }
Al Virofe479a52011-02-22 15:10:03 -05001594
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001595 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001597 /*
1598 * If it wasn't NUL, we know it was '/'. Skip that
1599 * slash, and continue until no more slashes.
1600 */
1601 do {
1602 len++;
1603 } while (unlikely(name[len] == '/'));
1604 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001605 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001606 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
Al Viroce57dfc2011-03-13 19:58:58 -04001608 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1609 if (err < 0)
1610 return err;
Al Virofe479a52011-02-22 15:10:03 -05001611
Al Viroce57dfc2011-03-13 19:58:58 -04001612 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001613 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001615 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001616 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001617 if (can_lookup(nd->inode))
1618 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001620 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 /* here ends the main loop */
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623last_component:
Al Virob3563792011-03-14 21:54:55 -04001624 nd->last = this;
1625 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001626 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 }
Al Viro951361f2011-03-04 14:44:37 -05001628 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 return err;
1630}
1631
Al Viro70e9b352011-03-05 21:12:22 -05001632static int path_init(int dfd, const char *name, unsigned int flags,
1633 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001635 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001636 int fput_needed;
1637 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
1639 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001640 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001642 if (flags & LOOKUP_ROOT) {
1643 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001644 if (*name) {
1645 if (!inode->i_op->lookup)
1646 return -ENOTDIR;
1647 retval = inode_permission(inode, MAY_EXEC);
1648 if (retval)
1649 return retval;
1650 }
Al Viro5b6ca022011-03-09 23:04:47 -05001651 nd->path = nd->root;
1652 nd->inode = inode;
1653 if (flags & LOOKUP_RCU) {
Andi Kleen962830d2012-05-08 13:32:02 +09301654 br_read_lock(&vfsmount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001655 rcu_read_lock();
1656 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1657 } else {
1658 path_get(&nd->path);
1659 }
1660 return 0;
1661 }
1662
Al Viro2a737872009-04-07 11:49:53 -04001663 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001666 if (flags & LOOKUP_RCU) {
Andi Kleen962830d2012-05-08 13:32:02 +09301667 br_read_lock(&vfsmount_lock);
Al Viroe41f7d42011-02-22 14:02:58 -05001668 rcu_read_lock();
1669 set_root_rcu(nd);
1670 } else {
1671 set_root(nd);
1672 path_get(&nd->root);
1673 }
Al Viro2a737872009-04-07 11:49:53 -04001674 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001675 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001676 if (flags & LOOKUP_RCU) {
1677 struct fs_struct *fs = current->fs;
1678 unsigned seq;
1679
Andi Kleen962830d2012-05-08 13:32:02 +09301680 br_read_lock(&vfsmount_lock);
Al Viroe41f7d42011-02-22 14:02:58 -05001681 rcu_read_lock();
1682
1683 do {
1684 seq = read_seqcount_begin(&fs->seq);
1685 nd->path = fs->pwd;
1686 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1687 } while (read_seqcount_retry(&fs->seq, seq));
1688 } else {
1689 get_fs_pwd(current->fs, &nd->path);
1690 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001691 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001692 struct dentry *dentry;
1693
Al Viro1abf0c72011-03-13 03:51:11 -04001694 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001695 retval = -EBADF;
1696 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001697 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001698
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001699 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001700
Al Virof52e0c12011-03-14 18:56:51 -04001701 if (*name) {
1702 retval = -ENOTDIR;
1703 if (!S_ISDIR(dentry->d_inode->i_mode))
1704 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001705
Al Viro4ad5abb2011-06-20 19:57:03 -04001706 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001707 if (retval)
1708 goto fput_fail;
1709 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001710
Jan Blunck5dd784d02008-02-14 19:34:38 -08001711 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001712 if (flags & LOOKUP_RCU) {
1713 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001714 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001715 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Andi Kleen962830d2012-05-08 13:32:02 +09301716 br_read_lock(&vfsmount_lock);
Al Viroe41f7d42011-02-22 14:02:58 -05001717 rcu_read_lock();
1718 } else {
1719 path_get(&file->f_path);
1720 fput_light(file, fput_needed);
1721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 }
Al Viroe41f7d42011-02-22 14:02:58 -05001723
Nick Piggin31e6b012011-01-07 17:49:52 +11001724 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001725 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001726
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001727fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001728 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001729out_fail:
1730 return retval;
1731}
1732
Al Virobd92d7f2011-03-14 19:54:59 -04001733static inline int lookup_last(struct nameidata *nd, struct path *path)
1734{
1735 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1736 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1737
1738 nd->flags &= ~LOOKUP_PARENT;
1739 return walk_component(nd, path, &nd->last, nd->last_type,
1740 nd->flags & LOOKUP_FOLLOW);
1741}
1742
Al Viro9b4a9b12009-04-07 11:44:16 -04001743/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001744static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001745 unsigned int flags, struct nameidata *nd)
1746{
Al Viro70e9b352011-03-05 21:12:22 -05001747 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001748 struct path path;
1749 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001750
1751 /*
1752 * Path walking is largely split up into 2 different synchronisation
1753 * schemes, rcu-walk and ref-walk (explained in
1754 * Documentation/filesystems/path-lookup.txt). These share much of the
1755 * path walk code, but some things particularly setup, cleanup, and
1756 * following mounts are sufficiently divergent that functions are
1757 * duplicated. Typically there is a function foo(), and its RCU
1758 * analogue, foo_rcu().
1759 *
1760 * -ECHILD is the error number of choice (just to avoid clashes) that
1761 * is returned if some aspect of an rcu-walk fails. Such an error must
1762 * be handled by restarting a traditional ref-walk (which will always
1763 * be able to complete).
1764 */
Al Virobd92d7f2011-03-14 19:54:59 -04001765 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001766
Al Virobd92d7f2011-03-14 19:54:59 -04001767 if (unlikely(err))
1768 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001769
1770 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001771 err = link_path_walk(name, nd);
1772
1773 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001774 err = lookup_last(nd, &path);
1775 while (err > 0) {
1776 void *cookie;
1777 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001778 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001779 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001780 if (!err)
1781 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001782 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001783 }
1784 }
Al Viroee0827c2011-02-21 23:38:09 -05001785
Al Viro9f1fafe2011-03-25 11:00:12 -04001786 if (!err)
1787 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001788
1789 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1790 if (!nd->inode->i_op->lookup) {
1791 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001792 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001793 }
1794 }
Al Viro16c2cd72011-02-22 15:50:10 -05001795
Al Viro70e9b352011-03-05 21:12:22 -05001796 if (base)
1797 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001798
Al Viro5b6ca022011-03-09 23:04:47 -05001799 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001800 path_put(&nd->root);
1801 nd->root.mnt = NULL;
1802 }
Al Virobd92d7f2011-03-14 19:54:59 -04001803 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001804}
Nick Piggin31e6b012011-01-07 17:49:52 +11001805
Al Viroee0827c2011-02-21 23:38:09 -05001806static int do_path_lookup(int dfd, const char *name,
1807 unsigned int flags, struct nameidata *nd)
1808{
1809 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1810 if (unlikely(retval == -ECHILD))
1811 retval = path_lookupat(dfd, name, flags, nd);
1812 if (unlikely(retval == -ESTALE))
1813 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001814
1815 if (likely(!retval)) {
1816 if (unlikely(!audit_dummy_context())) {
1817 if (nd->path.dentry && nd->inode)
1818 audit_inode(name, nd->path.dentry);
1819 }
1820 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001821 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822}
1823
Al Viroc9c6cac2011-02-16 15:15:47 -05001824int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001825{
Al Viroc9c6cac2011-02-16 15:15:47 -05001826 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001827}
1828
Al Virod1811462008-08-02 00:49:18 -04001829int kern_path(const char *name, unsigned int flags, struct path *path)
1830{
1831 struct nameidata nd;
1832 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1833 if (!res)
1834 *path = nd.path;
1835 return res;
1836}
1837
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001838/**
1839 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1840 * @dentry: pointer to dentry of the base directory
1841 * @mnt: pointer to vfs mount of the base directory
1842 * @name: pointer to file name
1843 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001844 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001845 */
1846int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1847 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001848 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001849{
Al Viroe0a01242011-06-27 17:00:37 -04001850 struct nameidata nd;
1851 int err;
1852 nd.root.dentry = dentry;
1853 nd.root.mnt = mnt;
1854 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001855 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001856 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1857 if (!err)
1858 *path = nd.path;
1859 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001860}
1861
James Morris057f6c02007-04-26 00:12:05 -07001862/*
1863 * Restricted form of lookup. Doesn't follow links, single-component only,
1864 * needs parent already locked. Doesn't follow mounts.
1865 * SMP-safe.
1866 */
Adrian Bunka244e162006-03-31 02:32:11 -08001867static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
Jan Blunck4ac91372008-02-14 19:34:32 -08001869 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870}
1871
Christoph Hellwigeead1912007-10-16 23:25:38 -07001872/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001873 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001874 * @name: pathname component to lookup
1875 * @base: base directory to lookup from
1876 * @len: maximum length @len should be interpreted to
1877 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001878 * Note that this routine is purely a helper for filesystem usage and should
1879 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001880 * nameidata argument is passed to the filesystem methods and a filesystem
1881 * using this helper needs to be prepared for that.
1882 */
James Morris057f6c02007-04-26 00:12:05 -07001883struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1884{
James Morris057f6c02007-04-26 00:12:05 -07001885 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001886 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02001887 int err;
James Morris057f6c02007-04-26 00:12:05 -07001888
David Woodhouse2f9092e2009-04-20 23:18:37 +01001889 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1890
Al Viro6a96ba52011-03-07 23:49:20 -05001891 this.name = name;
1892 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08001893 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05001894 if (!len)
1895 return ERR_PTR(-EACCES);
1896
Al Viro6a96ba52011-03-07 23:49:20 -05001897 while (len--) {
1898 c = *(const unsigned char *)name++;
1899 if (c == '/' || c == '\0')
1900 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05001901 }
Al Viro5a202bc2011-03-08 14:17:44 -05001902 /*
1903 * See if the low-level filesystem might want
1904 * to use its own hash..
1905 */
1906 if (base->d_flags & DCACHE_OP_HASH) {
1907 int err = base->d_op->d_hash(base, base->d_inode, &this);
1908 if (err < 0)
1909 return ERR_PTR(err);
1910 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001911
Miklos Szeredicda309d2012-03-26 12:54:21 +02001912 err = inode_permission(base->d_inode, MAY_EXEC);
1913 if (err)
1914 return ERR_PTR(err);
1915
Christoph Hellwig49705b72005-11-08 21:35:06 -08001916 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001917}
1918
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001919int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1920 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921{
Al Viro2d8f3032008-07-22 09:59:21 -04001922 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001923 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001926
1927 BUG_ON(flags & LOOKUP_PARENT);
1928
1929 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001931 if (!err)
1932 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 }
1934 return err;
1935}
1936
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001937int user_path_at(int dfd, const char __user *name, unsigned flags,
1938 struct path *path)
1939{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07001940 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001941}
1942
Al Viro2ad94ae2008-07-21 09:32:51 -04001943static int user_path_parent(int dfd, const char __user *path,
1944 struct nameidata *nd, char **name)
1945{
1946 char *s = getname(path);
1947 int error;
1948
1949 if (IS_ERR(s))
1950 return PTR_ERR(s);
1951
1952 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1953 if (error)
1954 putname(s);
1955 else
1956 *name = s;
1957
1958 return error;
1959}
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961/*
1962 * It's inline, so penalty for filesystems that don't use sticky bit is
1963 * minimal.
1964 */
1965static inline int check_sticky(struct inode *dir, struct inode *inode)
1966{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001967 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11001968
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 if (!(dir->i_mode & S_ISVTX))
1970 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001971 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001973 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08001975 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976}
1977
1978/*
1979 * Check whether we can remove a link victim from directory dir, check
1980 * whether the type of victim is right.
1981 * 1. We can't do it if dir is read-only (done in permission())
1982 * 2. We should have write and exec permissions on dir
1983 * 3. We can't remove anything from append-only dir
1984 * 4. We can't do anything with immutable dir (done in permission())
1985 * 5. If the sticky bit on dir is set we should either
1986 * a. be owner of dir, or
1987 * b. be owner of victim, or
1988 * c. have CAP_FOWNER capability
1989 * 6. If the victim is append-only or immutable we can't do antyhing with
1990 * links pointing to it.
1991 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1992 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1993 * 9. We can't remove a root or mountpoint.
1994 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1995 * nfs_async_unlink().
1996 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001997static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
1999 int error;
2000
2001 if (!victim->d_inode)
2002 return -ENOENT;
2003
2004 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05002005 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Al Virof419a2e2008-07-22 00:07:17 -04002007 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 if (error)
2009 return error;
2010 if (IS_APPEND(dir))
2011 return -EPERM;
2012 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002013 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 return -EPERM;
2015 if (isdir) {
2016 if (!S_ISDIR(victim->d_inode->i_mode))
2017 return -ENOTDIR;
2018 if (IS_ROOT(victim))
2019 return -EBUSY;
2020 } else if (S_ISDIR(victim->d_inode->i_mode))
2021 return -EISDIR;
2022 if (IS_DEADDIR(dir))
2023 return -ENOENT;
2024 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2025 return -EBUSY;
2026 return 0;
2027}
2028
2029/* Check whether we can create an object with dentry child in directory
2030 * dir.
2031 * 1. We can't do it if child already exists (open has special treatment for
2032 * this case, but since we are inlined it's OK)
2033 * 2. We can't do it if dir is read-only (done in permission())
2034 * 3. We should have write and exec permissions on dir
2035 * 4. We can't do it if dir is immutable (done in permission())
2036 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002037static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038{
2039 if (child->d_inode)
2040 return -EEXIST;
2041 if (IS_DEADDIR(dir))
2042 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002043 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044}
2045
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046/*
2047 * p1 and p2 should be directories on the same fs.
2048 */
2049struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2050{
2051 struct dentry *p;
2052
2053 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002054 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return NULL;
2056 }
2057
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002058 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002060 p = d_ancestor(p2, p1);
2061 if (p) {
2062 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2063 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2064 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
2066
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002067 p = d_ancestor(p1, p2);
2068 if (p) {
2069 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2070 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2071 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073
Ingo Molnarf2eace22006-07-03 00:25:05 -07002074 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2075 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 return NULL;
2077}
2078
2079void unlock_rename(struct dentry *p1, struct dentry *p2)
2080{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002081 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002083 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002084 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 }
2086}
2087
Al Viro4acdaf22011-07-26 01:42:34 -04002088int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 struct nameidata *nd)
2090{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002091 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
2093 if (error)
2094 return error;
2095
Al Viroacfa4382008-12-04 10:06:33 -05002096 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 return -EACCES; /* shouldn't it be ENOSYS? */
2098 mode &= S_IALLUGO;
2099 mode |= S_IFREG;
2100 error = security_inode_create(dir, dentry, mode);
2101 if (error)
2102 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002104 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002105 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return error;
2107}
2108
Al Viro73d049a2011-03-11 12:08:24 -05002109static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002111 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 struct inode *inode = dentry->d_inode;
2113 int error;
2114
Al Virobcda7652011-03-13 16:42:14 -04002115 /* O_PATH? */
2116 if (!acc_mode)
2117 return 0;
2118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 if (!inode)
2120 return -ENOENT;
2121
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002122 switch (inode->i_mode & S_IFMT) {
2123 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002125 case S_IFDIR:
2126 if (acc_mode & MAY_WRITE)
2127 return -EISDIR;
2128 break;
2129 case S_IFBLK:
2130 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002131 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002133 /*FALLTHRU*/
2134 case S_IFIFO:
2135 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002137 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002138 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002139
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002140 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002141 if (error)
2142 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002143
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 /*
2145 * An append-only file must be opened in append mode for writing.
2146 */
2147 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002148 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002149 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002151 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 }
2153
2154 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002155 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002156 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002158 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002159}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160
Jeff Laytone1181ee2010-12-07 16:19:50 -05002161static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002162{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002163 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002164 struct inode *inode = path->dentry->d_inode;
2165 int error = get_write_access(inode);
2166 if (error)
2167 return error;
2168 /*
2169 * Refuse to truncate files with mandatory locks held on them.
2170 */
2171 error = locks_verify_locked(inode);
2172 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002173 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002174 if (!error) {
2175 error = do_truncate(path->dentry, 0,
2176 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002177 filp);
Al Viro7715b522009-12-16 03:54:00 -05002178 }
2179 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002180 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181}
2182
Dave Hansend57999e2008-02-15 14:37:27 -08002183static inline int open_to_namei_flags(int flag)
2184{
Al Viro8a5e9292011-06-25 19:15:54 -04002185 if ((flag & O_ACCMODE) == 3)
2186 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002187 return flag;
2188}
2189
Nick Piggin31e6b012011-01-07 17:49:52 +11002190/*
Al Virofe2d35f2011-03-05 22:58:25 -05002191 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002192 */
Al Virofb1cc552009-12-24 01:58:28 -05002193static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002194 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002195{
Al Viroa1e28032009-12-24 02:12:06 -05002196 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002197 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002198 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002199 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002200 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002201 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002202 struct file *filp;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002203 struct inode *inode;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002204 int symlink_ok = 0;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002205 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2206 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002207 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002208
Al Viroc3e380b2011-02-23 13:39:45 -05002209 nd->flags &= ~LOOKUP_PARENT;
2210 nd->flags |= op->intent;
2211
Al Viro1f36f772009-12-26 10:56:19 -05002212 switch (nd->last_type) {
2213 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002214 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002215 error = handle_dots(nd, nd->last_type);
2216 if (error)
2217 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002218 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002219 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002220 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002221 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002222 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002223 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002224 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002225 error = -EISDIR;
2226 goto exit;
2227 }
2228 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002229 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002230 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002231 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002232 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002233 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002234 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002235 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002236
Al Viroca344a892011-03-09 00:36:45 -05002237 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002238 if (nd->last.name[nd->last.len])
2239 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002240 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2241 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002242 /* we _can_ be in RCU mode here */
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002243 error = lookup_fast(nd, &nd->last, path, &inode);
2244 if (unlikely(error)) {
2245 if (error < 0)
2246 goto exit;
2247
2248 error = lookup_slow(nd, &nd->last, path);
2249 if (error < 0)
2250 goto exit;
2251
2252 inode = path->dentry->d_inode;
2253 }
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002254 goto finish_lookup;
Al Virofe2d35f2011-03-05 22:58:25 -05002255 }
2256
2257 /* create side of things */
Al Viroa3fbbde2011-11-07 21:21:26 +00002258 /*
2259 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2260 * cleared when we got to the last component we are about to look up
2261 */
Al Viro9f1fafe2011-03-25 11:00:12 -04002262 error = complete_walk(nd);
2263 if (error)
2264 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002265
2266 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002267 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002268 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002269 if (nd->last.name[nd->last.len])
2270 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002271
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002272retry_lookup:
Al Viroa1e28032009-12-24 02:12:06 -05002273 mutex_lock(&dir->d_inode->i_mutex);
2274
Al Viro6c0d46c2011-03-09 00:59:59 -05002275 dentry = lookup_hash(nd);
2276 error = PTR_ERR(dentry);
2277 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002278 mutex_unlock(&dir->d_inode->i_mutex);
2279 goto exit;
2280 }
2281
Al Viro6c0d46c2011-03-09 00:59:59 -05002282 path->dentry = dentry;
2283 path->mnt = nd->path.mnt;
2284
Al Virofb1cc552009-12-24 01:58:28 -05002285 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002286 if (!dentry->d_inode) {
Al Viroa218d0f2011-11-21 14:59:34 -05002287 umode_t mode = op->mode;
Al Viro6c0d46c2011-03-09 00:59:59 -05002288 if (!IS_POSIXACL(dir->d_inode))
2289 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002290 /*
2291 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002292 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002293 * the time when the file is created and when
2294 * a permanent write count is taken through
2295 * the 'struct file' in nameidata_to_filp().
2296 */
2297 error = mnt_want_write(nd->path.mnt);
2298 if (error)
2299 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002300 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002301 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002302 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002303 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002304 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002305 error = security_path_mknod(&nd->path, dentry, mode, 0);
2306 if (error)
2307 goto exit_mutex_unlock;
2308 error = vfs_create(dir->d_inode, dentry, mode, nd);
2309 if (error)
2310 goto exit_mutex_unlock;
2311 mutex_unlock(&dir->d_inode->i_mutex);
2312 dput(nd->path.dentry);
2313 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002314 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002315 }
2316
2317 /*
2318 * It already exists.
2319 */
2320 mutex_unlock(&dir->d_inode->i_mutex);
2321 audit_inode(pathname, path->dentry);
2322
2323 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002324 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002325 goto exit_dput;
2326
David Howells9875cf82011-01-14 18:45:21 +00002327 error = follow_managed(path, nd->flags);
2328 if (error < 0)
2329 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002330
Al Viroa3fbbde2011-11-07 21:21:26 +00002331 if (error)
2332 nd->flags |= LOOKUP_JUMPED;
2333
Miklos Szeredidecf3402012-05-21 17:30:08 +02002334 BUG_ON(nd->flags & LOOKUP_RCU);
2335 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002336finish_lookup:
2337 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002338 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002339 if (!inode) {
2340 path_to_nameidata(path, nd);
2341 goto exit;
2342 }
Al Viro9e67f362009-12-26 07:04:50 -05002343
Miklos Szeredid45ea862012-05-21 17:30:09 +02002344 if (should_follow_link(inode, !symlink_ok)) {
2345 if (nd->flags & LOOKUP_RCU) {
2346 if (unlikely(unlazy_walk(nd, path->dentry))) {
2347 error = -ECHILD;
2348 goto exit;
2349 }
2350 }
2351 BUG_ON(inode != path->dentry->d_inode);
Al Virofb1cc552009-12-24 01:58:28 -05002352 return NULL;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002353 }
Al Virofb1cc552009-12-24 01:58:28 -05002354
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002355 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2356 path_to_nameidata(path, nd);
2357 } else {
2358 save_parent.dentry = nd->path.dentry;
2359 save_parent.mnt = mntget(path->mnt);
2360 nd->path.dentry = path->dentry;
2361
2362 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002363 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002364 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2365 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002366 if (error) {
2367 path_put(&save_parent);
Miklos Szeredi097b1802012-03-06 13:56:33 +01002368 return ERR_PTR(error);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002369 }
Al Virofb1cc552009-12-24 01:58:28 -05002370 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002371 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002372 goto exit;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002373 error = -ENOTDIR;
2374 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2375 goto exit;
Miklos Szeredid7fdd7f2012-05-21 17:30:13 +02002376 audit_inode(pathname, nd->path.dentry);
Al Viro67ee3ad2009-12-26 07:01:01 -05002377ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002378 if (!S_ISREG(nd->inode->i_mode))
2379 will_truncate = 0;
2380
Al Viro0f9d1a12011-03-09 00:13:14 -05002381 if (will_truncate) {
2382 error = mnt_want_write(nd->path.mnt);
2383 if (error)
2384 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002385 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002386 }
Al Viroca344a892011-03-09 00:36:45 -05002387common:
Al Virobcda7652011-03-13 16:42:14 -04002388 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002389 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002390 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002391 filp = nameidata_to_filp(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002392 if (filp == ERR_PTR(-EOPENSTALE) && save_parent.dentry && !retried) {
2393 BUG_ON(save_parent.dentry != dir);
2394 path_put(&nd->path);
2395 nd->path = save_parent;
2396 nd->inode = dir->d_inode;
2397 save_parent.mnt = NULL;
2398 save_parent.dentry = NULL;
2399 if (want_write) {
2400 mnt_drop_write(nd->path.mnt);
2401 want_write = 0;
2402 }
2403 retried = true;
2404 goto retry_lookup;
2405 }
Al Viro0f9d1a12011-03-09 00:13:14 -05002406 if (!IS_ERR(filp)) {
2407 error = ima_file_check(filp, op->acc_mode);
2408 if (error) {
2409 fput(filp);
2410 filp = ERR_PTR(error);
2411 }
2412 }
2413 if (!IS_ERR(filp)) {
2414 if (will_truncate) {
2415 error = handle_truncate(filp);
2416 if (error) {
2417 fput(filp);
2418 filp = ERR_PTR(error);
2419 }
2420 }
2421 }
Al Viroca344a892011-03-09 00:36:45 -05002422out:
2423 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002424 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002425 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002426 terminate_walk(nd);
Al Virofb1cc552009-12-24 01:58:28 -05002427 return filp;
2428
2429exit_mutex_unlock:
2430 mutex_unlock(&dir->d_inode->i_mutex);
2431exit_dput:
2432 path_put_conditional(path, nd);
2433exit:
Al Viroca344a892011-03-09 00:36:45 -05002434 filp = ERR_PTR(error);
2435 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002436}
2437
Al Viro13aab422011-02-23 17:54:08 -05002438static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002439 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440{
Al Virofe2d35f2011-03-05 22:58:25 -05002441 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002442 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002443 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002444 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002445
2446 filp = get_empty_filp();
2447 if (!filp)
2448 return ERR_PTR(-ENFILE);
2449
Al Viro47c805d2011-02-23 17:44:09 -05002450 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002451 nd->intent.open.file = filp;
2452 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2453 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002454
Al Viro73d049a2011-03-11 12:08:24 -05002455 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002456 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002457 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002458
Al Virofe2d35f2011-03-05 22:58:25 -05002459 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002460 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002461 if (unlikely(error))
2462 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Al Viro73d049a2011-03-11 12:08:24 -05002464 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002465 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002466 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002467 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002468 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002469 path_put_conditional(&path, nd);
2470 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002471 filp = ERR_PTR(-ELOOP);
2472 break;
2473 }
Al Viro73d049a2011-03-11 12:08:24 -05002474 nd->flags |= LOOKUP_PARENT;
2475 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002476 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002477 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002478 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002479 else
Al Viro73d049a2011-03-11 12:08:24 -05002480 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002481 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002482 }
Al Viro10fa8e62009-12-26 07:09:49 -05002483out:
Al Viro73d049a2011-03-11 12:08:24 -05002484 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2485 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002486 if (base)
2487 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002488 release_open_intent(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002489 if (filp == ERR_PTR(-EOPENSTALE)) {
2490 if (flags & LOOKUP_RCU)
2491 filp = ERR_PTR(-ECHILD);
2492 else
2493 filp = ERR_PTR(-ESTALE);
2494 }
Al Viro10fa8e62009-12-26 07:09:49 -05002495 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
Nick Piggin31e6b012011-01-07 17:49:52 +11002497out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002498 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002499 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500}
2501
Al Viro13aab422011-02-23 17:54:08 -05002502struct file *do_filp_open(int dfd, const char *pathname,
2503 const struct open_flags *op, int flags)
2504{
Al Viro73d049a2011-03-11 12:08:24 -05002505 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002506 struct file *filp;
2507
Al Viro73d049a2011-03-11 12:08:24 -05002508 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002509 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002510 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002511 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002512 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002513 return filp;
2514}
2515
Al Viro73d049a2011-03-11 12:08:24 -05002516struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2517 const char *name, const struct open_flags *op, int flags)
2518{
2519 struct nameidata nd;
2520 struct file *file;
2521
2522 nd.root.mnt = mnt;
2523 nd.root.dentry = dentry;
2524
2525 flags |= LOOKUP_ROOT;
2526
Al Virobcda7652011-03-13 16:42:14 -04002527 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002528 return ERR_PTR(-ELOOP);
2529
2530 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2531 if (unlikely(file == ERR_PTR(-ECHILD)))
2532 file = path_openat(-1, name, &nd, op, flags);
2533 if (unlikely(file == ERR_PTR(-ESTALE)))
2534 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2535 return file;
2536}
2537
Al Viroed75e952011-06-27 16:53:43 -04002538struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002540 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002541 struct nameidata nd;
2542 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2543 if (error)
2544 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002546 /*
2547 * Yucky last component or no last component at all?
2548 * (foo/., foo/.., /////)
2549 */
Al Viroed75e952011-06-27 16:53:43 -04002550 if (nd.last_type != LAST_NORM)
2551 goto out;
2552 nd.flags &= ~LOOKUP_PARENT;
2553 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2554 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002555
2556 /*
2557 * Do the final lookup.
2558 */
Al Viroed75e952011-06-27 16:53:43 -04002559 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2560 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 if (IS_ERR(dentry))
2562 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002563
Al Viroe9baf6e2008-05-15 04:49:12 -04002564 if (dentry->d_inode)
2565 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002566 /*
2567 * Special case - lookup gave negative, but... we had foo/bar/
2568 * From the vfs_mknod() POV we just have a negative dentry -
2569 * all is fine. Let's be bastards - you had / on the end, you've
2570 * been asking for (non-existent) directory. -ENOENT for you.
2571 */
Al Viroed75e952011-06-27 16:53:43 -04002572 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002573 dput(dentry);
2574 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002575 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002576 }
Al Viroed75e952011-06-27 16:53:43 -04002577 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002579eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002581 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582fail:
Al Viroed75e952011-06-27 16:53:43 -04002583 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2584out:
2585 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 return dentry;
2587}
Al Virodae6ad82011-06-26 11:50:15 -04002588EXPORT_SYMBOL(kern_path_create);
2589
2590struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2591{
2592 char *tmp = getname(pathname);
2593 struct dentry *res;
2594 if (IS_ERR(tmp))
2595 return ERR_CAST(tmp);
2596 res = kern_path_create(dfd, tmp, path, is_dir);
2597 putname(tmp);
2598 return res;
2599}
2600EXPORT_SYMBOL(user_path_create);
2601
Al Viro1a67aaf2011-07-26 01:52:52 -04002602int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002604 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
2606 if (error)
2607 return error;
2608
Eric W. Biederman975d6b32011-11-13 12:16:43 -08002609 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 return -EPERM;
2611
Al Viroacfa4382008-12-04 10:06:33 -05002612 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 return -EPERM;
2614
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002615 error = devcgroup_inode_mknod(mode, dev);
2616 if (error)
2617 return error;
2618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 error = security_inode_mknod(dir, dentry, mode, dev);
2620 if (error)
2621 return error;
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002624 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002625 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 return error;
2627}
2628
Al Virof69aac02011-07-26 04:31:40 -04002629static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08002630{
2631 switch (mode & S_IFMT) {
2632 case S_IFREG:
2633 case S_IFCHR:
2634 case S_IFBLK:
2635 case S_IFIFO:
2636 case S_IFSOCK:
2637 case 0: /* zero mode translates to S_IFREG */
2638 return 0;
2639 case S_IFDIR:
2640 return -EPERM;
2641 default:
2642 return -EINVAL;
2643 }
2644}
2645
Al Viro8208a222011-07-25 17:32:17 -04002646SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002647 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648{
Al Viro2ad94ae2008-07-21 09:32:51 -04002649 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002650 struct path path;
2651 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
2653 if (S_ISDIR(mode))
2654 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Al Virodae6ad82011-06-26 11:50:15 -04002656 dentry = user_path_create(dfd, filename, &path, 0);
2657 if (IS_ERR(dentry))
2658 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002659
Al Virodae6ad82011-06-26 11:50:15 -04002660 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002661 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002662 error = may_mknod(mode);
2663 if (error)
2664 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002665 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002666 if (error)
2667 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002668 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002669 if (error)
2670 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002671 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002673 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 break;
2675 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002676 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 new_decode_dev(dev));
2678 break;
2679 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002680 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002683out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002684 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002685out_dput:
2686 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002687 mutex_unlock(&path.dentry->d_inode->i_mutex);
2688 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689
2690 return error;
2691}
2692
Al Viro8208a222011-07-25 17:32:17 -04002693SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002694{
2695 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2696}
2697
Al Viro18bb1db2011-07-26 01:41:39 -04002698int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002700 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05002701 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
2703 if (error)
2704 return error;
2705
Al Viroacfa4382008-12-04 10:06:33 -05002706 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 return -EPERM;
2708
2709 mode &= (S_IRWXUGO|S_ISVTX);
2710 error = security_inode_mkdir(dir, dentry, mode);
2711 if (error)
2712 return error;
2713
Al Viro8de52772012-02-06 12:45:27 -05002714 if (max_links && dir->i_nlink >= max_links)
2715 return -EMLINK;
2716
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002718 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002719 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 return error;
2721}
2722
Al Viroa218d0f2011-11-21 14:59:34 -05002723SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724{
Dave Hansen6902d922006-09-30 23:29:01 -07002725 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002726 struct path path;
2727 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Al Virodae6ad82011-06-26 11:50:15 -04002729 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002730 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002731 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002732
Al Virodae6ad82011-06-26 11:50:15 -04002733 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002734 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002735 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002736 if (error)
2737 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002738 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002739 if (error)
2740 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002741 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002742out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002743 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002744out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002745 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002746 mutex_unlock(&path.dentry->d_inode->i_mutex);
2747 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 return error;
2749}
2750
Al Viroa218d0f2011-11-21 14:59:34 -05002751SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002752{
2753 return sys_mkdirat(AT_FDCWD, pathname, mode);
2754}
2755
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756/*
Sage Weila71905f2011-05-24 13:06:08 -07002757 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05002758 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07002759 * dentry, and if that is true (possibly after pruning the dcache),
2760 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 *
2762 * A low-level filesystem can, if it choses, legally
2763 * do a
2764 *
2765 * if (!d_unhashed(dentry))
2766 * return -EBUSY;
2767 *
2768 * if it cannot handle the case of removing a directory
2769 * that is still in use by something else..
2770 */
2771void dentry_unhash(struct dentry *dentry)
2772{
Vasily Averindc168422006-12-06 20:37:07 -08002773 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002775 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 __d_drop(dentry);
2777 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778}
2779
2780int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2781{
2782 int error = may_delete(dir, dentry, 1);
2783
2784 if (error)
2785 return error;
2786
Al Viroacfa4382008-12-04 10:06:33 -05002787 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 return -EPERM;
2789
Al Viro1d2ef592011-09-14 18:55:41 +01002790 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002791 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792
Sage Weil912dbc12011-05-24 13:06:11 -07002793 error = -EBUSY;
2794 if (d_mountpoint(dentry))
2795 goto out;
2796
2797 error = security_inode_rmdir(dir, dentry);
2798 if (error)
2799 goto out;
2800
Sage Weil3cebde22011-05-29 21:20:59 -07002801 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002802 error = dir->i_op->rmdir(dir, dentry);
2803 if (error)
2804 goto out;
2805
2806 dentry->d_inode->i_flags |= S_DEAD;
2807 dont_mount(dentry);
2808
2809out:
2810 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01002811 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002812 if (!error)
2813 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 return error;
2815}
2816
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002817static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818{
2819 int error = 0;
2820 char * name;
2821 struct dentry *dentry;
2822 struct nameidata nd;
2823
Al Viro2ad94ae2008-07-21 09:32:51 -04002824 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002826 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827
2828 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002829 case LAST_DOTDOT:
2830 error = -ENOTEMPTY;
2831 goto exit1;
2832 case LAST_DOT:
2833 error = -EINVAL;
2834 goto exit1;
2835 case LAST_ROOT:
2836 error = -EBUSY;
2837 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002839
2840 nd.flags &= ~LOOKUP_PARENT;
2841
Jan Blunck4ac91372008-02-14 19:34:32 -08002842 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002843 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002845 if (IS_ERR(dentry))
2846 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002847 if (!dentry->d_inode) {
2848 error = -ENOENT;
2849 goto exit3;
2850 }
Dave Hansen06227532008-02-15 14:37:34 -08002851 error = mnt_want_write(nd.path.mnt);
2852 if (error)
2853 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002854 error = security_path_rmdir(&nd.path, dentry);
2855 if (error)
2856 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002857 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002858exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002859 mnt_drop_write(nd.path.mnt);
2860exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002861 dput(dentry);
2862exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002863 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002865 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 putname(name);
2867 return error;
2868}
2869
Heiko Carstens3cdad422009-01-14 14:14:22 +01002870SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002871{
2872 return do_rmdir(AT_FDCWD, pathname);
2873}
2874
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875int vfs_unlink(struct inode *dir, struct dentry *dentry)
2876{
2877 int error = may_delete(dir, dentry, 0);
2878
2879 if (error)
2880 return error;
2881
Al Viroacfa4382008-12-04 10:06:33 -05002882 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 return -EPERM;
2884
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002885 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 if (d_mountpoint(dentry))
2887 error = -EBUSY;
2888 else {
2889 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002890 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002892 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002893 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002896 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897
2898 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2899 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002900 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 }
Robert Love0eeca282005-07-12 17:06:03 -04002903
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 return error;
2905}
2906
2907/*
2908 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002909 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 * writeout happening, and we don't want to prevent access to the directory
2911 * while waiting on the I/O.
2912 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002913static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914{
Al Viro2ad94ae2008-07-21 09:32:51 -04002915 int error;
2916 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 struct dentry *dentry;
2918 struct nameidata nd;
2919 struct inode *inode = NULL;
2920
Al Viro2ad94ae2008-07-21 09:32:51 -04002921 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002923 return error;
2924
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 error = -EISDIR;
2926 if (nd.last_type != LAST_NORM)
2927 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002928
2929 nd.flags &= ~LOOKUP_PARENT;
2930
Jan Blunck4ac91372008-02-14 19:34:32 -08002931 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002932 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 error = PTR_ERR(dentry);
2934 if (!IS_ERR(dentry)) {
2935 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002936 if (nd.last.name[nd.last.len])
2937 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002939 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002940 goto slashes;
2941 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002942 error = mnt_want_write(nd.path.mnt);
2943 if (error)
2944 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002945 error = security_path_unlink(&nd.path, dentry);
2946 if (error)
2947 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002948 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002949exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002950 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 exit2:
2952 dput(dentry);
2953 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002954 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 if (inode)
2956 iput(inode); /* truncate the inode here */
2957exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002958 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 putname(name);
2960 return error;
2961
2962slashes:
2963 error = !dentry->d_inode ? -ENOENT :
2964 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2965 goto exit2;
2966}
2967
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002968SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002969{
2970 if ((flag & ~AT_REMOVEDIR) != 0)
2971 return -EINVAL;
2972
2973 if (flag & AT_REMOVEDIR)
2974 return do_rmdir(dfd, pathname);
2975
2976 return do_unlinkat(dfd, pathname);
2977}
2978
Heiko Carstens3480b252009-01-14 14:14:16 +01002979SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002980{
2981 return do_unlinkat(AT_FDCWD, pathname);
2982}
2983
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002984int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002986 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
2988 if (error)
2989 return error;
2990
Al Viroacfa4382008-12-04 10:06:33 -05002991 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 return -EPERM;
2993
2994 error = security_inode_symlink(dir, dentry, oldname);
2995 if (error)
2996 return error;
2997
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002999 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003000 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 return error;
3002}
3003
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003004SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3005 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006{
Al Viro2ad94ae2008-07-21 09:32:51 -04003007 int error;
3008 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003009 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003010 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011
3012 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003013 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003015
Al Virodae6ad82011-06-26 11:50:15 -04003016 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07003017 error = PTR_ERR(dentry);
3018 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003019 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003020
Al Virodae6ad82011-06-26 11:50:15 -04003021 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003022 if (error)
3023 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003024 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003025 if (error)
3026 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003027 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003028out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003029 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003030out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003031 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003032 mutex_unlock(&path.dentry->d_inode->i_mutex);
3033 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07003034out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 putname(from);
3036 return error;
3037}
3038
Heiko Carstens3480b252009-01-14 14:14:16 +01003039SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003040{
3041 return sys_symlinkat(oldname, AT_FDCWD, newname);
3042}
3043
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3045{
3046 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003047 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 int error;
3049
3050 if (!inode)
3051 return -ENOENT;
3052
Miklos Szeredia95164d2008-07-30 15:08:48 +02003053 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 if (error)
3055 return error;
3056
3057 if (dir->i_sb != inode->i_sb)
3058 return -EXDEV;
3059
3060 /*
3061 * A link to an append-only or immutable file cannot be created.
3062 */
3063 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3064 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003065 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003067 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 return -EPERM;
3069
3070 error = security_inode_link(old_dentry, dir, new_dentry);
3071 if (error)
3072 return error;
3073
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003074 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303075 /* Make sure we don't allow creating hardlink to an unlinked file */
3076 if (inode->i_nlink == 0)
3077 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003078 else if (max_links && inode->i_nlink >= max_links)
3079 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303080 else
3081 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003082 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003083 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003084 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 return error;
3086}
3087
3088/*
3089 * Hardlinks are often used in delicate situations. We avoid
3090 * security-related surprises by not following symlinks on the
3091 * newname. --KAB
3092 *
3093 * We don't follow them on the oldname either to be compatible
3094 * with linux 2.0, and to avoid hard-linking to directories
3095 * and other special files. --ADM
3096 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003097SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3098 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099{
3100 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003101 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303102 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303105 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003106 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303107 /*
3108 * To use null names we require CAP_DAC_READ_SEARCH
3109 * This ensures that not everyone will be able to create
3110 * handlink using the passed filedescriptor.
3111 */
3112 if (flags & AT_EMPTY_PATH) {
3113 if (!capable(CAP_DAC_READ_SEARCH))
3114 return -ENOENT;
3115 how = LOOKUP_EMPTY;
3116 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003117
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303118 if (flags & AT_SYMLINK_FOLLOW)
3119 how |= LOOKUP_FOLLOW;
3120
3121 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003123 return error;
3124
Al Virodae6ad82011-06-26 11:50:15 -04003125 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003127 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003128 goto out;
3129
3130 error = -EXDEV;
3131 if (old_path.mnt != new_path.mnt)
3132 goto out_dput;
3133 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003134 if (error)
3135 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003136 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003137 if (error)
3138 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003139 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003140out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003141 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003142out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003143 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003144 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3145 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146out:
Al Viro2d8f3032008-07-22 09:59:21 -04003147 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148
3149 return error;
3150}
3151
Heiko Carstens3480b252009-01-14 14:14:16 +01003152SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003153{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003154 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003155}
3156
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157/*
3158 * The worst of all namespace operations - renaming directory. "Perverted"
3159 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3160 * Problems:
3161 * a) we can get into loop creation. Check is done in is_subdir().
3162 * b) race potential - two innocent renames can create a loop together.
3163 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003164 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 * story.
3166 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003167 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 * whether the target exists). Solution: try to be smart with locking
3169 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003170 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 * move will be locked. Thus we can rank directories by the tree
3172 * (ancestors first) and rank all non-directories after them.
3173 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003174 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 * HOWEVER, it relies on the assumption that any object with ->lookup()
3176 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3177 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003178 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003179 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003181 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 * locking].
3183 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003184static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3185 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186{
3187 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003188 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003189 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190
3191 /*
3192 * If we are going to change the parent - check write permissions,
3193 * we'll need to flip '..'.
3194 */
3195 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003196 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 if (error)
3198 return error;
3199 }
3200
3201 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3202 if (error)
3203 return error;
3204
Al Viro1d2ef592011-09-14 18:55:41 +01003205 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003206 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003207 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003208
3209 error = -EBUSY;
3210 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3211 goto out;
3212
Al Viro8de52772012-02-06 12:45:27 -05003213 error = -EMLINK;
3214 if (max_links && !target && new_dir != old_dir &&
3215 new_dir->i_nlink >= max_links)
3216 goto out;
3217
Sage Weil3cebde22011-05-29 21:20:59 -07003218 if (target)
3219 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003220 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3221 if (error)
3222 goto out;
3223
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003225 target->i_flags |= S_DEAD;
3226 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 }
Sage Weil9055cba2011-05-24 13:06:12 -07003228out:
3229 if (target)
3230 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003231 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003232 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003233 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3234 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235 return error;
3236}
3237
Adrian Bunk75c96f82005-05-05 16:16:09 -07003238static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3239 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240{
Sage Weil51892bb2011-05-24 13:06:13 -07003241 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 int error;
3243
3244 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3245 if (error)
3246 return error;
3247
3248 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003250 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003251
3252 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003254 goto out;
3255
3256 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3257 if (error)
3258 goto out;
3259
3260 if (target)
3261 dont_mount(new_dentry);
3262 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3263 d_move(old_dentry, new_dentry);
3264out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003266 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267 dput(new_dentry);
3268 return error;
3269}
3270
3271int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3272 struct inode *new_dir, struct dentry *new_dentry)
3273{
3274 int error;
3275 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003276 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277
3278 if (old_dentry->d_inode == new_dentry->d_inode)
3279 return 0;
3280
3281 error = may_delete(old_dir, old_dentry, is_dir);
3282 if (error)
3283 return error;
3284
3285 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003286 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 else
3288 error = may_delete(new_dir, new_dentry, is_dir);
3289 if (error)
3290 return error;
3291
Al Viroacfa4382008-12-04 10:06:33 -05003292 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 return -EPERM;
3294
Robert Love0eeca282005-07-12 17:06:03 -04003295 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3296
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 if (is_dir)
3298 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3299 else
3300 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003301 if (!error)
3302 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003303 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003304 fsnotify_oldname_free(old_name);
3305
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 return error;
3307}
3308
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003309SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3310 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311{
Al Viro2ad94ae2008-07-21 09:32:51 -04003312 struct dentry *old_dir, *new_dir;
3313 struct dentry *old_dentry, *new_dentry;
3314 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003316 char *from;
3317 char *to;
3318 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319
Al Viro2ad94ae2008-07-21 09:32:51 -04003320 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 if (error)
3322 goto exit;
3323
Al Viro2ad94ae2008-07-21 09:32:51 -04003324 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 if (error)
3326 goto exit1;
3327
3328 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003329 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 goto exit2;
3331
Jan Blunck4ac91372008-02-14 19:34:32 -08003332 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 error = -EBUSY;
3334 if (oldnd.last_type != LAST_NORM)
3335 goto exit2;
3336
Jan Blunck4ac91372008-02-14 19:34:32 -08003337 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 if (newnd.last_type != LAST_NORM)
3339 goto exit2;
3340
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003341 oldnd.flags &= ~LOOKUP_PARENT;
3342 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003343 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003344
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 trap = lock_rename(new_dir, old_dir);
3346
Christoph Hellwig49705b72005-11-08 21:35:06 -08003347 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 error = PTR_ERR(old_dentry);
3349 if (IS_ERR(old_dentry))
3350 goto exit3;
3351 /* source must exist */
3352 error = -ENOENT;
3353 if (!old_dentry->d_inode)
3354 goto exit4;
3355 /* unless the source is a directory trailing slashes give -ENOTDIR */
3356 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3357 error = -ENOTDIR;
3358 if (oldnd.last.name[oldnd.last.len])
3359 goto exit4;
3360 if (newnd.last.name[newnd.last.len])
3361 goto exit4;
3362 }
3363 /* source should not be ancestor of target */
3364 error = -EINVAL;
3365 if (old_dentry == trap)
3366 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003367 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 error = PTR_ERR(new_dentry);
3369 if (IS_ERR(new_dentry))
3370 goto exit4;
3371 /* target should not be an ancestor of source */
3372 error = -ENOTEMPTY;
3373 if (new_dentry == trap)
3374 goto exit5;
3375
Dave Hansen9079b1e2008-02-15 14:37:49 -08003376 error = mnt_want_write(oldnd.path.mnt);
3377 if (error)
3378 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003379 error = security_path_rename(&oldnd.path, old_dentry,
3380 &newnd.path, new_dentry);
3381 if (error)
3382 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 error = vfs_rename(old_dir->d_inode, old_dentry,
3384 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003385exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003386 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387exit5:
3388 dput(new_dentry);
3389exit4:
3390 dput(old_dentry);
3391exit3:
3392 unlock_rename(new_dir, old_dir);
3393exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003394 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003395 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003397 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003399exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 return error;
3401}
3402
Heiko Carstensa26eab22009-01-14 14:14:17 +01003403SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003404{
3405 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3406}
3407
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3409{
3410 int len;
3411
3412 len = PTR_ERR(link);
3413 if (IS_ERR(link))
3414 goto out;
3415
3416 len = strlen(link);
3417 if (len > (unsigned) buflen)
3418 len = buflen;
3419 if (copy_to_user(buffer, link, len))
3420 len = -EFAULT;
3421out:
3422 return len;
3423}
3424
3425/*
3426 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3427 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3428 * using) it for any given inode is up to filesystem.
3429 */
3430int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3431{
3432 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003433 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003434 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003435
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003437 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003438 if (IS_ERR(cookie))
3439 return PTR_ERR(cookie);
3440
3441 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3442 if (dentry->d_inode->i_op->put_link)
3443 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3444 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445}
3446
3447int vfs_follow_link(struct nameidata *nd, const char *link)
3448{
3449 return __vfs_follow_link(nd, link);
3450}
3451
3452/* get the link contents into pagecache */
3453static char *page_getlink(struct dentry * dentry, struct page **ppage)
3454{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003455 char *kaddr;
3456 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003458 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003460 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003462 kaddr = kmap(page);
3463 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3464 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465}
3466
3467int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3468{
3469 struct page *page = NULL;
3470 char *s = page_getlink(dentry, &page);
3471 int res = vfs_readlink(dentry,buffer,buflen,s);
3472 if (page) {
3473 kunmap(page);
3474 page_cache_release(page);
3475 }
3476 return res;
3477}
3478
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003479void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003481 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003483 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484}
3485
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003486void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003488 struct page *page = cookie;
3489
3490 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 kunmap(page);
3492 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 }
3494}
3495
Nick Piggin54566b22009-01-04 12:00:53 -08003496/*
3497 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3498 */
3499int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500{
3501 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003502 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003503 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003504 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003506 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3507 if (nofs)
3508 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509
NeilBrown7e53cac2006-03-25 03:07:57 -08003510retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003511 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003512 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003514 goto fail;
3515
Cong Wange8e3c3d2011-11-25 23:14:27 +08003516 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08003518 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07003519
3520 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3521 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 if (err < 0)
3523 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003524 if (err < len-1)
3525 goto retry;
3526
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 mark_inode_dirty(inode);
3528 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529fail:
3530 return err;
3531}
3532
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003533int page_symlink(struct inode *inode, const char *symname, int len)
3534{
3535 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003536 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003537}
3538
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003539const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540 .readlink = generic_readlink,
3541 .follow_link = page_follow_link_light,
3542 .put_link = page_put_link,
3543};
3544
Al Viro2d8f3032008-07-22 09:59:21 -04003545EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003546EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547EXPORT_SYMBOL(follow_down);
3548EXPORT_SYMBOL(follow_up);
3549EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3550EXPORT_SYMBOL(getname);
3551EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552EXPORT_SYMBOL(lookup_one_len);
3553EXPORT_SYMBOL(page_follow_link_light);
3554EXPORT_SYMBOL(page_put_link);
3555EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003556EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557EXPORT_SYMBOL(page_symlink);
3558EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003559EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003560EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003561EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562EXPORT_SYMBOL(unlock_rename);
3563EXPORT_SYMBOL(vfs_create);
3564EXPORT_SYMBOL(vfs_follow_link);
3565EXPORT_SYMBOL(vfs_link);
3566EXPORT_SYMBOL(vfs_mkdir);
3567EXPORT_SYMBOL(vfs_mknod);
3568EXPORT_SYMBOL(generic_permission);
3569EXPORT_SYMBOL(vfs_readlink);
3570EXPORT_SYMBOL(vfs_rename);
3571EXPORT_SYMBOL(vfs_rmdir);
3572EXPORT_SYMBOL(vfs_symlink);
3573EXPORT_SYMBOL(vfs_unlink);
3574EXPORT_SYMBOL(dentry_unhash);
3575EXPORT_SYMBOL(generic_readlink);