blob: 0b3138de2a3b79fd8d1bf1843e405506f343f605 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070035#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37
Eric Parise81e3f42009-12-04 15:47:36 -050038#include "internal.h"
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030074 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
Al Virof52e0c12011-03-14 18:56:51 -0400140static char *getname_flags(const char __user * filename, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
Al Virof52e0c12011-03-14 18:56:51 -0400151 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
152 __putname(tmp);
153 result = ERR_PTR(retval);
154 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
156 }
157 audit_getname(result);
158 return result;
159}
160
Al Virof52e0c12011-03-14 18:56:51 -0400161char *getname(const char __user * filename)
162{
163 return getname_flags(filename, 0);
164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166#ifdef CONFIG_AUDITSYSCALL
167void putname(const char *name)
168{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400169 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 audit_putname(name);
171 else
172 __putname(name);
173}
174EXPORT_SYMBOL(putname);
175#endif
176
Linus Torvaldse77819e2011-07-22 19:30:19 -0700177static int check_acl(struct inode *inode, int mask)
178{
Linus Torvalds84635d62011-07-25 22:47:03 -0700179#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700180 struct posix_acl *acl;
181
Linus Torvaldse77819e2011-07-22 19:30:19 -0700182 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400183 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
184 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700185 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400186 /* no ->get_acl() calls in RCU mode... */
187 if (acl == ACL_NOT_CACHED)
188 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300189 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700190 }
191
192 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
193
194 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200195 * A filesystem can force a ACL callback by just never filling the
196 * ACL cache. But normally you'd fill the cache either at inode
197 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700198 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200199 * If the filesystem doesn't have a get_acl() function at all, we'll
200 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700201 */
202 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200203 if (inode->i_op->get_acl) {
204 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
205 if (IS_ERR(acl))
206 return PTR_ERR(acl);
207 } else {
208 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
209 return -EAGAIN;
210 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700211 }
212
213 if (acl) {
214 int error = posix_acl_permission(inode, acl, mask);
215 posix_acl_release(acl);
216 return error;
217 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700218#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700219
220 return -EAGAIN;
221}
222
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700223/*
224 * This does basic POSIX ACL permission checking
225 */
Al Viro7e401452011-06-20 19:12:17 -0400226static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700227{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700228 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229
Al Viro9c2c7032011-06-20 19:06:22 -0400230 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700231
Serge E. Hallyne795b712011-03-23 16:43:25 -0700232 if (current_user_ns() != inode_userns(inode))
233 goto other_perms;
234
Linus Torvalds14067ff2011-07-25 19:55:52 -0700235 if (likely(current_fsuid() == inode->i_uid))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700236 mode >>= 6;
237 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400239 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100240 if (error != -EAGAIN)
241 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700242 }
243
244 if (in_group_p(inode->i_gid))
245 mode >>= 3;
246 }
247
Serge E. Hallyne795b712011-03-23 16:43:25 -0700248other_perms:
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700249 /*
250 * If the DACs are ok we don't need any capability check.
251 */
Al Viro9c2c7032011-06-20 19:06:22 -0400252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700253 return 0;
254 return -EACCES;
255}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100258 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * @inode: inode to check access rights for
260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100265 * are used for other things.
266 *
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 */
Al Viro2830ba72011-06-20 19:16:29 -0400271int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700273 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700276 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 */
Al Viro7e401452011-06-20 19:12:17 -0400278 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700279 if (ret != -EACCES)
280 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Al Virod594e7e2011-06-20 19:55:42 -0400282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
285 return 0;
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
288 return 0;
289 return -EACCES;
290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 /*
292 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 */
Al Virod594e7e2011-06-20 19:55:42 -0400296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 return 0;
299
300 /*
301 * Searching includes executable on directories, else just read.
302 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400304 if (mask == MAY_READ)
Serge E. Hallyne795b712011-03-23 16:43:25 -0700305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return 0;
307
308 return -EACCES;
309}
310
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700311/*
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
316 */
317static inline int do_inode_permission(struct inode *inode, int mask)
318{
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
322
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
327 }
328 return generic_permission(inode, mask);
329}
330
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200331/**
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
335 *
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
340 */
Al Virof419a2e2008-07-22 00:07:17 -0400341int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Al Viroe6305c42008-07-15 21:03:57 -0400343 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700345 if (unlikely(mask & MAY_WRITE)) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700346 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 /*
349 * Nobody gets write access to a read-only fs.
350 */
351 if (IS_RDONLY(inode) &&
352 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
353 return -EROFS;
354
355 /*
356 * Nobody gets write access to an immutable file.
357 */
358 if (IS_IMMUTABLE(inode))
359 return -EACCES;
360 }
361
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700362 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (retval)
364 return retval;
365
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700366 retval = devcgroup_inode_permission(inode, mask);
367 if (retval)
368 return retval;
369
Eric Parisd09ca732010-07-23 11:43:57 -0400370 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Al Virof4d6ff82011-06-19 13:14:21 -0400373/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800374 * path_get - get a reference to a path
375 * @path: path to get the reference to
376 *
377 * Given a path increment the reference count to the dentry and the vfsmount.
378 */
379void path_get(struct path *path)
380{
381 mntget(path->mnt);
382 dget(path->dentry);
383}
384EXPORT_SYMBOL(path_get);
385
386/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800387 * path_put - put a reference to a path
388 * @path: path to put the reference to
389 *
390 * Given a path decrement the reference count to the dentry and the vfsmount.
391 */
392void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Jan Blunck1d957f92008-02-14 19:34:35 -0800394 dput(path->dentry);
395 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
Jan Blunck1d957f92008-02-14 19:34:35 -0800397EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Al Viro19660af2011-03-25 10:32:48 -0400399/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100400 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400401 * Documentation/filesystems/path-lookup.txt). In situations when we can't
402 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
403 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
404 * mode. Refcounts are grabbed at the last known good point before rcu-walk
405 * got stuck, so ref-walk may continue from there. If this is not successful
406 * (eg. a seqcount has changed), then failure is returned and it's up to caller
407 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100408 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100409
410/**
Al Viro19660af2011-03-25 10:32:48 -0400411 * unlazy_walk - try to switch to ref-walk mode.
412 * @nd: nameidata pathwalk data
413 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800414 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100415 *
Al Viro19660af2011-03-25 10:32:48 -0400416 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
417 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
418 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100419 */
Al Viro19660af2011-03-25 10:32:48 -0400420static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100421{
422 struct fs_struct *fs = current->fs;
423 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500424 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100425
426 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500427 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
428 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100429 spin_lock(&fs->lock);
430 if (nd->root.mnt != fs->root.mnt ||
431 nd->root.dentry != fs->root.dentry)
432 goto err_root;
433 }
434 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400435 if (!dentry) {
436 if (!__d_rcu_to_refcount(parent, nd->seq))
437 goto err_parent;
438 BUG_ON(nd->inode != parent->d_inode);
439 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400440 if (dentry->d_parent != parent)
441 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400442 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
443 if (!__d_rcu_to_refcount(dentry, nd->seq))
444 goto err_child;
445 /*
446 * If the sequence check on the child dentry passed, then
447 * the child has not been removed from its parent. This
448 * means the parent dentry must be valid and able to take
449 * a reference at this point.
450 */
451 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
452 BUG_ON(!parent->d_count);
453 parent->d_count++;
454 spin_unlock(&dentry->d_lock);
455 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100456 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500457 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100458 path_get(&nd->root);
459 spin_unlock(&fs->lock);
460 }
461 mntget(nd->path.mnt);
462
463 rcu_read_unlock();
464 br_read_unlock(vfsmount_lock);
465 nd->flags &= ~LOOKUP_RCU;
466 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400467
468err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100469 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400470err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100471 spin_unlock(&parent->d_lock);
472err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500473 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100474 spin_unlock(&fs->lock);
475 return -ECHILD;
476}
477
Nick Piggin31e6b012011-01-07 17:49:52 +1100478/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700479 * release_open_intent - free up open intent resources
480 * @nd: pointer to nameidata
481 */
482void release_open_intent(struct nameidata *nd)
483{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800484 struct file *file = nd->intent.open.file;
485
486 if (file && !IS_ERR(file)) {
487 if (file->f_path.dentry == NULL)
488 put_filp(file);
489 else
490 fput(file);
491 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700492}
493
Al Virof60aef72011-02-15 01:35:28 -0500494static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100495{
Al Virof60aef72011-02-15 01:35:28 -0500496 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100497}
498
Al Viro9f1fafe2011-03-25 11:00:12 -0400499/**
500 * complete_walk - successful completion of path walk
501 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500502 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400503 * If we had been in RCU mode, drop out of it and legitimize nd->path.
504 * Revalidate the final result, unless we'd already done that during
505 * the path walk or the filesystem doesn't ask for it. Return 0 on
506 * success, -error on failure. In case of failure caller does not
507 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500508 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400509static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500510{
Al Viro16c2cd72011-02-22 15:50:10 -0500511 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500512 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500513
Al Viro9f1fafe2011-03-25 11:00:12 -0400514 if (nd->flags & LOOKUP_RCU) {
515 nd->flags &= ~LOOKUP_RCU;
516 if (!(nd->flags & LOOKUP_ROOT))
517 nd->root.mnt = NULL;
518 spin_lock(&dentry->d_lock);
519 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
520 spin_unlock(&dentry->d_lock);
521 rcu_read_unlock();
522 br_read_unlock(vfsmount_lock);
523 return -ECHILD;
524 }
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
527 mntget(nd->path.mnt);
528 rcu_read_unlock();
529 br_read_unlock(vfsmount_lock);
530 }
531
Al Viro16c2cd72011-02-22 15:50:10 -0500532 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500533 return 0;
534
Al Viro16c2cd72011-02-22 15:50:10 -0500535 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
536 return 0;
537
538 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
539 return 0;
540
541 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100542 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500543 if (status > 0)
544 return 0;
545
Al Viro16c2cd72011-02-22 15:50:10 -0500546 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500547 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500548
Al Viro9f1fafe2011-03-25 11:00:12 -0400549 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500550 return status;
551}
552
Al Viro2a737872009-04-07 11:49:53 -0400553static __always_inline void set_root(struct nameidata *nd)
554{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200555 if (!nd->root.mnt)
556 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400557}
558
Al Viro6de88d72009-08-09 01:41:57 +0400559static int link_path_walk(const char *, struct nameidata *);
560
Nick Piggin31e6b012011-01-07 17:49:52 +1100561static __always_inline void set_root_rcu(struct nameidata *nd)
562{
563 if (!nd->root.mnt) {
564 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100565 unsigned seq;
566
567 do {
568 seq = read_seqcount_begin(&fs->seq);
569 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700570 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100571 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100572 }
573}
574
Arjan van de Venf1662352006-01-14 13:21:31 -0800575static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Nick Piggin31e6b012011-01-07 17:49:52 +1100577 int ret;
578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (IS_ERR(link))
580 goto fail;
581
582 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400583 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800584 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400585 nd->path = nd->root;
586 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500587 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100589 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100590
Nick Piggin31e6b012011-01-07 17:49:52 +1100591 ret = link_path_walk(link, nd);
592 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800594 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 return PTR_ERR(link);
596}
597
Jan Blunck1d957f92008-02-14 19:34:35 -0800598static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700599{
600 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800601 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700602 mntput(path->mnt);
603}
604
Nick Piggin7b9337a2011-01-14 08:42:43 +0000605static inline void path_to_nameidata(const struct path *path,
606 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700607{
Nick Piggin31e6b012011-01-07 17:49:52 +1100608 if (!(nd->flags & LOOKUP_RCU)) {
609 dput(nd->path.dentry);
610 if (nd->path.mnt != path->mnt)
611 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800612 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100613 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800614 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700615}
616
Al Viro574197e2011-03-14 22:20:34 -0400617static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
618{
619 struct inode *inode = link->dentry->d_inode;
620 if (!IS_ERR(cookie) && inode->i_op->put_link)
621 inode->i_op->put_link(link->dentry, nd, cookie);
622 path_put(link);
623}
624
Al Virodef4af32009-12-26 08:37:05 -0500625static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400626follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800627{
628 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000629 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800630
Al Viro844a3912011-02-15 00:38:26 -0500631 BUG_ON(nd->flags & LOOKUP_RCU);
632
Al Viro0e794582011-03-16 02:45:02 -0400633 if (link->mnt == nd->path.mnt)
634 mntget(link->mnt);
635
Al Viro574197e2011-03-14 22:20:34 -0400636 if (unlikely(current->total_link_count >= 40)) {
637 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400638 path_put(&nd->path);
639 return -ELOOP;
640 }
641 cond_resched();
642 current->total_link_count++;
643
Nick Piggin7b9337a2011-01-14 08:42:43 +0000644 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800645 nd_set_link(nd, NULL);
646
Al Viro36f3b4f2011-02-22 21:24:38 -0500647 error = security_inode_follow_link(link->dentry, nd);
648 if (error) {
649 *p = ERR_PTR(error); /* no ->put_link(), please */
650 path_put(&nd->path);
651 return error;
652 }
653
Al Viro86acdca12009-12-22 23:45:11 -0500654 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500655 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
656 error = PTR_ERR(*p);
657 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800658 char *s = nd_get_link(nd);
659 error = 0;
660 if (s)
661 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400662 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500663 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400664 nd->inode = nd->path.dentry->d_inode;
665 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400666 /* stepped on a _really_ weird one */
667 path_put(&nd->path);
668 error = -ELOOP;
669 }
670 }
Ian Kent051d3812006-03-27 01:14:53 -0800671 }
Ian Kent051d3812006-03-27 01:14:53 -0800672 return error;
673}
674
Nick Piggin31e6b012011-01-07 17:49:52 +1100675static int follow_up_rcu(struct path *path)
676{
677 struct vfsmount *parent;
678 struct dentry *mountpoint;
679
680 parent = path->mnt->mnt_parent;
681 if (parent == path->mnt)
682 return 0;
683 mountpoint = path->mnt->mnt_mountpoint;
684 path->dentry = mountpoint;
685 path->mnt = parent;
686 return 1;
687}
688
Al Virobab77eb2009-04-18 03:26:48 -0400689int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 struct vfsmount *parent;
692 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000693
694 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400695 parent = path->mnt->mnt_parent;
696 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000697 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 return 0;
699 }
700 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400701 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000702 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400703 dput(path->dentry);
704 path->dentry = mountpoint;
705 mntput(path->mnt);
706 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 return 1;
708}
709
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100710/*
David Howells9875cf82011-01-14 18:45:21 +0000711 * Perform an automount
712 * - return -EISDIR to tell follow_managed() to stop and return the path we
713 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 */
David Howells9875cf82011-01-14 18:45:21 +0000715static int follow_automount(struct path *path, unsigned flags,
716 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100717{
David Howells9875cf82011-01-14 18:45:21 +0000718 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000719 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100720
David Howells9875cf82011-01-14 18:45:21 +0000721 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
722 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700723
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200724 /* We don't want to mount if someone's just doing a stat -
725 * unless they're stat'ing a directory and appended a '/' to
726 * the name.
727 *
728 * We do, however, want to mount if someone wants to open or
729 * create a file of any type under the mountpoint, wants to
730 * traverse through the mountpoint or wants to open the
731 * mounted directory. Also, autofs may mark negative dentries
732 * as being automount points. These will need the attentions
733 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000734 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200735 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700736 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200737 path->dentry->d_inode)
738 return -EISDIR;
739
David Howells9875cf82011-01-14 18:45:21 +0000740 current->total_link_count++;
741 if (current->total_link_count >= 40)
742 return -ELOOP;
743
744 mnt = path->dentry->d_op->d_automount(path);
745 if (IS_ERR(mnt)) {
746 /*
747 * The filesystem is allowed to return -EISDIR here to indicate
748 * it doesn't want to automount. For instance, autofs would do
749 * this so that its userspace daemon can mount on this dentry.
750 *
751 * However, we can only permit this if it's a terminal point in
752 * the path being looked up; if it wasn't then the remainder of
753 * the path is inaccessible and we should say so.
754 */
Al Viro49084c32011-06-25 21:59:52 -0400755 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000756 return -EREMOTE;
757 return PTR_ERR(mnt);
758 }
David Howellsea5b7782011-01-14 19:10:03 +0000759
David Howells9875cf82011-01-14 18:45:21 +0000760 if (!mnt) /* mount collision */
761 return 0;
762
Al Viro8aef1882011-06-16 15:10:06 +0100763 if (!*need_mntput) {
764 /* lock_mount() may release path->mnt on error */
765 mntget(path->mnt);
766 *need_mntput = true;
767 }
Al Viro19a167a2011-01-17 01:35:23 -0500768 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000769
David Howellsea5b7782011-01-14 19:10:03 +0000770 switch (err) {
771 case -EBUSY:
772 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500773 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000774 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100775 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000776 path->mnt = mnt;
777 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000778 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500779 default:
780 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000781 }
Al Viro19a167a2011-01-17 01:35:23 -0500782
David Howells9875cf82011-01-14 18:45:21 +0000783}
784
785/*
786 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000787 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000788 * - Flagged as mountpoint
789 * - Flagged as automount point
790 *
791 * This may only be called in refwalk mode.
792 *
793 * Serialization is taken care of in namespace.c
794 */
795static int follow_managed(struct path *path, unsigned flags)
796{
Al Viro8aef1882011-06-16 15:10:06 +0100797 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000798 unsigned managed;
799 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100800 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000801
802 /* Given that we're not holding a lock here, we retain the value in a
803 * local variable for each dentry as we look at it so that we don't see
804 * the components of that value change under us */
805 while (managed = ACCESS_ONCE(path->dentry->d_flags),
806 managed &= DCACHE_MANAGED_DENTRY,
807 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000808 /* Allow the filesystem to manage the transit without i_mutex
809 * being held. */
810 if (managed & DCACHE_MANAGE_TRANSIT) {
811 BUG_ON(!path->dentry->d_op);
812 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400813 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000814 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100815 break;
David Howellscc53ce52011-01-14 18:45:26 +0000816 }
817
David Howells9875cf82011-01-14 18:45:21 +0000818 /* Transit to a mounted filesystem. */
819 if (managed & DCACHE_MOUNTED) {
820 struct vfsmount *mounted = lookup_mnt(path);
821 if (mounted) {
822 dput(path->dentry);
823 if (need_mntput)
824 mntput(path->mnt);
825 path->mnt = mounted;
826 path->dentry = dget(mounted->mnt_root);
827 need_mntput = true;
828 continue;
829 }
830
831 /* Something is mounted on this dentry in another
832 * namespace and/or whatever was mounted there in this
833 * namespace got unmounted before we managed to get the
834 * vfsmount_lock */
835 }
836
837 /* Handle an automount point */
838 if (managed & DCACHE_NEED_AUTOMOUNT) {
839 ret = follow_automount(path, flags, &need_mntput);
840 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100841 break;
David Howells9875cf82011-01-14 18:45:21 +0000842 continue;
843 }
844
845 /* We didn't change the current path point */
846 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 }
Al Viro8aef1882011-06-16 15:10:06 +0100848
849 if (need_mntput && path->mnt == mnt)
850 mntput(path->mnt);
851 if (ret == -EISDIR)
852 ret = 0;
853 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854}
855
David Howellscc53ce52011-01-14 18:45:26 +0000856int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
858 struct vfsmount *mounted;
859
Al Viro1c755af2009-04-18 14:06:57 -0400860 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400862 dput(path->dentry);
863 mntput(path->mnt);
864 path->mnt = mounted;
865 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 return 1;
867 }
868 return 0;
869}
870
Ian Kent62a73752011-03-25 01:51:02 +0800871static inline bool managed_dentry_might_block(struct dentry *dentry)
872{
873 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
874 dentry->d_op->d_manage(dentry, true) < 0);
875}
876
David Howells9875cf82011-01-14 18:45:21 +0000877/*
Al Viro287548e2011-05-27 06:50:06 -0400878 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
879 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000880 */
881static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400882 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000883{
Ian Kent62a73752011-03-25 01:51:02 +0800884 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +0000885 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800886 /*
887 * Don't forget we might have a non-mountpoint managed dentry
888 * that wants to block transit.
889 */
Al Viro287548e2011-05-27 06:50:06 -0400890 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000891 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800892
893 if (!d_mountpoint(path->dentry))
894 break;
895
David Howells9875cf82011-01-14 18:45:21 +0000896 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
897 if (!mounted)
898 break;
899 path->mnt = mounted;
900 path->dentry = mounted->mnt_root;
901 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700902 /*
903 * Update the inode too. We don't need to re-check the
904 * dentry sequence number here after this d_inode read,
905 * because a mount-point is always pinned.
906 */
907 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000908 }
David Howells9875cf82011-01-14 18:45:21 +0000909 return true;
910}
911
Al Virodea39372011-05-27 06:53:39 -0400912static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400913{
Al Virodea39372011-05-27 06:53:39 -0400914 while (d_mountpoint(nd->path.dentry)) {
Al Viro287548e2011-05-27 06:50:06 -0400915 struct vfsmount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400916 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400917 if (!mounted)
918 break;
Al Virodea39372011-05-27 06:53:39 -0400919 nd->path.mnt = mounted;
920 nd->path.dentry = mounted->mnt_root;
921 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400922 }
923}
924
Nick Piggin31e6b012011-01-07 17:49:52 +1100925static int follow_dotdot_rcu(struct nameidata *nd)
926{
Nick Piggin31e6b012011-01-07 17:49:52 +1100927 set_root_rcu(nd);
928
David Howells9875cf82011-01-14 18:45:21 +0000929 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100930 if (nd->path.dentry == nd->root.dentry &&
931 nd->path.mnt == nd->root.mnt) {
932 break;
933 }
934 if (nd->path.dentry != nd->path.mnt->mnt_root) {
935 struct dentry *old = nd->path.dentry;
936 struct dentry *parent = old->d_parent;
937 unsigned seq;
938
939 seq = read_seqcount_begin(&parent->d_seq);
940 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500941 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100942 nd->path.dentry = parent;
943 nd->seq = seq;
944 break;
945 }
946 if (!follow_up_rcu(&nd->path))
947 break;
948 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100949 }
Al Virodea39372011-05-27 06:53:39 -0400950 follow_mount_rcu(nd);
951 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100952 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500953
954failed:
955 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500956 if (!(nd->flags & LOOKUP_ROOT))
957 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500958 rcu_read_unlock();
959 br_read_unlock(vfsmount_lock);
960 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100961}
962
David Howells9875cf82011-01-14 18:45:21 +0000963/*
David Howellscc53ce52011-01-14 18:45:26 +0000964 * Follow down to the covering mount currently visible to userspace. At each
965 * point, the filesystem owning that dentry may be queried as to whether the
966 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000967 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400968int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000969{
970 unsigned managed;
971 int ret;
972
973 while (managed = ACCESS_ONCE(path->dentry->d_flags),
974 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
975 /* Allow the filesystem to manage the transit without i_mutex
976 * being held.
977 *
978 * We indicate to the filesystem if someone is trying to mount
979 * something here. This gives autofs the chance to deny anyone
980 * other than its daemon the right to mount on its
981 * superstructure.
982 *
983 * The filesystem may sleep at this point.
984 */
985 if (managed & DCACHE_MANAGE_TRANSIT) {
986 BUG_ON(!path->dentry->d_op);
987 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000988 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400989 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 }
993
994 /* Transit to a mounted filesystem. */
995 if (managed & DCACHE_MOUNTED) {
996 struct vfsmount *mounted = lookup_mnt(path);
997 if (!mounted)
998 break;
999 dput(path->dentry);
1000 mntput(path->mnt);
1001 path->mnt = mounted;
1002 path->dentry = dget(mounted->mnt_root);
1003 continue;
1004 }
1005
1006 /* Don't handle automount points here */
1007 break;
1008 }
1009 return 0;
1010}
1011
1012/*
David Howells9875cf82011-01-14 18:45:21 +00001013 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1014 */
1015static void follow_mount(struct path *path)
1016{
1017 while (d_mountpoint(path->dentry)) {
1018 struct vfsmount *mounted = lookup_mnt(path);
1019 if (!mounted)
1020 break;
1021 dput(path->dentry);
1022 mntput(path->mnt);
1023 path->mnt = mounted;
1024 path->dentry = dget(mounted->mnt_root);
1025 }
1026}
1027
Nick Piggin31e6b012011-01-07 17:49:52 +11001028static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029{
Al Viro2a737872009-04-07 11:49:53 -04001030 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001033 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Al Viro2a737872009-04-07 11:49:53 -04001035 if (nd->path.dentry == nd->root.dentry &&
1036 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 break;
1038 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001039 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001040 /* rare case of legitimate dget_parent()... */
1041 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 dput(old);
1043 break;
1044 }
Al Viro3088dd72010-01-30 15:47:29 -05001045 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
Al Viro79ed0222009-04-18 13:59:41 -04001048 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001049 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001053 * Allocate a dentry with name and parent, and perform a parent
1054 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1055 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1056 * have verified that no child exists while under i_mutex.
1057 */
1058static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1059 struct qstr *name, struct nameidata *nd)
1060{
1061 struct inode *inode = parent->d_inode;
1062 struct dentry *dentry;
1063 struct dentry *old;
1064
1065 /* Don't create child dentry for a dead directory. */
1066 if (unlikely(IS_DEADDIR(inode)))
1067 return ERR_PTR(-ENOENT);
1068
1069 dentry = d_alloc(parent, name);
1070 if (unlikely(!dentry))
1071 return ERR_PTR(-ENOMEM);
1072
1073 old = inode->i_op->lookup(inode, dentry, nd);
1074 if (unlikely(old)) {
1075 dput(dentry);
1076 dentry = old;
1077 }
1078 return dentry;
1079}
1080
1081/*
Josef Bacik44396f42011-05-31 11:58:49 -04001082 * We already have a dentry, but require a lookup to be performed on the parent
1083 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1084 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1085 * child exists while under i_mutex.
1086 */
1087static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1088 struct nameidata *nd)
1089{
1090 struct inode *inode = parent->d_inode;
1091 struct dentry *old;
1092
1093 /* Don't create child dentry for a dead directory. */
1094 if (unlikely(IS_DEADDIR(inode)))
1095 return ERR_PTR(-ENOENT);
1096
1097 old = inode->i_op->lookup(inode, dentry, nd);
1098 if (unlikely(old)) {
1099 dput(dentry);
1100 dentry = old;
1101 }
1102 return dentry;
1103}
1104
1105/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 * It's more convoluted than I'd like it to be, but... it's still fairly
1107 * small and for now I'd prefer to have fast path as straight as possible.
1108 * It _is_ time-critical.
1109 */
1110static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001111 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
Jan Blunck4ac91372008-02-14 19:34:32 -08001113 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001114 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001115 int need_reval = 1;
1116 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001117 int err;
1118
Al Viro3cac2602009-08-13 18:27:43 +04001119 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001120 * Rename seqlock is not required here because in the off chance
1121 * of a false negative due to a concurrent rename, we're going to
1122 * do the non-racy lookup, below.
1123 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001124 if (nd->flags & LOOKUP_RCU) {
1125 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001126 *inode = nd->inode;
1127 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001128 if (!dentry)
1129 goto unlazy;
1130
Nick Piggin31e6b012011-01-07 17:49:52 +11001131 /* Memory barrier in read_seqcount_begin of child is enough */
1132 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1133 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001134 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001135
Al Viro24643082011-02-15 01:26:22 -05001136 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001137 status = d_revalidate(dentry, nd);
1138 if (unlikely(status <= 0)) {
1139 if (status != -ECHILD)
1140 need_reval = 0;
1141 goto unlazy;
1142 }
Al Viro24643082011-02-15 01:26:22 -05001143 }
Josef Bacik44396f42011-05-31 11:58:49 -04001144 if (unlikely(d_need_lookup(dentry)))
1145 goto unlazy;
Nick Piggin31e6b012011-01-07 17:49:52 +11001146 path->mnt = mnt;
1147 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001148 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1149 goto unlazy;
1150 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1151 goto unlazy;
1152 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001153unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001154 if (unlazy_walk(nd, dentry))
1155 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001156 } else {
1157 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001158 }
Al Viro5a18fff2011-03-11 04:44:53 -05001159
Josef Bacik44396f42011-05-31 11:58:49 -04001160 if (dentry && unlikely(d_need_lookup(dentry))) {
1161 dput(dentry);
1162 dentry = NULL;
1163 }
Al Viro5a18fff2011-03-11 04:44:53 -05001164retry:
1165 if (unlikely(!dentry)) {
1166 struct inode *dir = parent->d_inode;
1167 BUG_ON(nd->inode != dir);
1168
1169 mutex_lock(&dir->i_mutex);
1170 dentry = d_lookup(parent, name);
1171 if (likely(!dentry)) {
1172 dentry = d_alloc_and_lookup(parent, name, nd);
1173 if (IS_ERR(dentry)) {
1174 mutex_unlock(&dir->i_mutex);
1175 return PTR_ERR(dentry);
1176 }
1177 /* known good */
1178 need_reval = 0;
1179 status = 1;
Josef Bacik44396f42011-05-31 11:58:49 -04001180 } else if (unlikely(d_need_lookup(dentry))) {
1181 dentry = d_inode_lookup(parent, dentry, nd);
1182 if (IS_ERR(dentry)) {
1183 mutex_unlock(&dir->i_mutex);
1184 return PTR_ERR(dentry);
1185 }
1186 /* known good */
1187 need_reval = 0;
1188 status = 1;
Al Viro5a18fff2011-03-11 04:44:53 -05001189 }
1190 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001191 }
Al Viro5a18fff2011-03-11 04:44:53 -05001192 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1193 status = d_revalidate(dentry, nd);
1194 if (unlikely(status <= 0)) {
1195 if (status < 0) {
1196 dput(dentry);
1197 return status;
1198 }
1199 if (!d_invalidate(dentry)) {
1200 dput(dentry);
1201 dentry = NULL;
1202 need_reval = 1;
1203 goto retry;
1204 }
1205 }
1206
David Howells9875cf82011-01-14 18:45:21 +00001207 path->mnt = mnt;
1208 path->dentry = dentry;
1209 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001210 if (unlikely(err < 0)) {
1211 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001212 return err;
Ian Kent89312212011-01-18 12:06:10 +08001213 }
David Howells9875cf82011-01-14 18:45:21 +00001214 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
1217
Al Viro52094c82011-02-21 21:34:47 -05001218static inline int may_lookup(struct nameidata *nd)
1219{
1220 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001221 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001222 if (err != -ECHILD)
1223 return err;
Al Viro19660af2011-03-25 10:32:48 -04001224 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001225 return -ECHILD;
1226 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001227 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001228}
1229
Al Viro9856fa12011-03-04 14:22:06 -05001230static inline int handle_dots(struct nameidata *nd, int type)
1231{
1232 if (type == LAST_DOTDOT) {
1233 if (nd->flags & LOOKUP_RCU) {
1234 if (follow_dotdot_rcu(nd))
1235 return -ECHILD;
1236 } else
1237 follow_dotdot(nd);
1238 }
1239 return 0;
1240}
1241
Al Viro951361f2011-03-04 14:44:37 -05001242static void terminate_walk(struct nameidata *nd)
1243{
1244 if (!(nd->flags & LOOKUP_RCU)) {
1245 path_put(&nd->path);
1246 } else {
1247 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001248 if (!(nd->flags & LOOKUP_ROOT))
1249 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001250 rcu_read_unlock();
1251 br_read_unlock(vfsmount_lock);
1252 }
1253}
1254
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001255/*
1256 * Do we need to follow links? We _really_ want to be able
1257 * to do this check without having to look at inode->i_op,
1258 * so we keep a cache of "no, this doesn't need follow_link"
1259 * for the common case.
1260 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001261static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001262{
1263 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1264 if (likely(inode->i_op->follow_link))
1265 return follow;
1266
1267 /* This gets set once for the inode lifetime */
1268 spin_lock(&inode->i_lock);
1269 inode->i_opflags |= IOP_NOFOLLOW;
1270 spin_unlock(&inode->i_lock);
1271 }
1272 return 0;
1273}
1274
Al Viroce57dfc2011-03-13 19:58:58 -04001275static inline int walk_component(struct nameidata *nd, struct path *path,
1276 struct qstr *name, int type, int follow)
1277{
1278 struct inode *inode;
1279 int err;
1280 /*
1281 * "." and ".." are special - ".." especially so because it has
1282 * to be able to know about the current root directory and
1283 * parent relationships.
1284 */
1285 if (unlikely(type != LAST_NORM))
1286 return handle_dots(nd, type);
1287 err = do_lookup(nd, name, path, &inode);
1288 if (unlikely(err)) {
1289 terminate_walk(nd);
1290 return err;
1291 }
1292 if (!inode) {
1293 path_to_nameidata(path, nd);
1294 terminate_walk(nd);
1295 return -ENOENT;
1296 }
Linus Torvalds7813b942011-08-07 09:53:20 -07001297 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001298 if (nd->flags & LOOKUP_RCU) {
1299 if (unlikely(unlazy_walk(nd, path->dentry))) {
1300 terminate_walk(nd);
1301 return -ECHILD;
1302 }
1303 }
Al Viroce57dfc2011-03-13 19:58:58 -04001304 BUG_ON(inode != path->dentry->d_inode);
1305 return 1;
1306 }
1307 path_to_nameidata(path, nd);
1308 nd->inode = inode;
1309 return 0;
1310}
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312/*
Al Virob3563792011-03-14 21:54:55 -04001313 * This limits recursive symlink follows to 8, while
1314 * limiting consecutive symlinks to 40.
1315 *
1316 * Without that kind of total limit, nasty chains of consecutive
1317 * symlinks can cause almost arbitrarily long lookups.
1318 */
1319static inline int nested_symlink(struct path *path, struct nameidata *nd)
1320{
1321 int res;
1322
Al Virob3563792011-03-14 21:54:55 -04001323 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1324 path_put_conditional(path, nd);
1325 path_put(&nd->path);
1326 return -ELOOP;
1327 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001328 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001329
1330 nd->depth++;
1331 current->link_count++;
1332
1333 do {
1334 struct path link = *path;
1335 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001336
1337 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001338 if (!res)
1339 res = walk_component(nd, path, &nd->last,
1340 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001341 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001342 } while (res > 0);
1343
1344 current->link_count--;
1345 nd->depth--;
1346 return res;
1347}
1348
1349/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001350 * We really don't want to look at inode->i_op->lookup
1351 * when we don't have to. So we keep a cache bit in
1352 * the inode ->i_opflags field that says "yes, we can
1353 * do lookup on this inode".
1354 */
1355static inline int can_lookup(struct inode *inode)
1356{
1357 if (likely(inode->i_opflags & IOP_LOOKUP))
1358 return 1;
1359 if (likely(!inode->i_op->lookup))
1360 return 0;
1361
1362 /* We do this once for the lifetime of the inode */
1363 spin_lock(&inode->i_lock);
1364 inode->i_opflags |= IOP_LOOKUP;
1365 spin_unlock(&inode->i_lock);
1366 return 1;
1367}
1368
1369/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001371 * This is the basic name resolution function, turning a pathname into
1372 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001374 * Returns 0 and nd will have valid dentry and mnt on success.
1375 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 */
Al Viro6de88d72009-08-09 01:41:57 +04001377static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378{
1379 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382 while (*name=='/')
1383 name++;
1384 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001385 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 /* At this point we know we have a real path component. */
1388 for(;;) {
1389 unsigned long hash;
1390 struct qstr this;
1391 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001392 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Al Viro52094c82011-02-21 21:34:47 -05001394 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 if (err)
1396 break;
1397
1398 this.name = name;
1399 c = *(const unsigned char *)name;
1400
1401 hash = init_name_hash();
1402 do {
1403 name++;
1404 hash = partial_name_hash(c, hash);
1405 c = *(const unsigned char *)name;
1406 } while (c && (c != '/'));
1407 this.len = name - (const char *) this.name;
1408 this.hash = end_name_hash(hash);
1409
Al Virofe479a52011-02-22 15:10:03 -05001410 type = LAST_NORM;
1411 if (this.name[0] == '.') switch (this.len) {
1412 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001413 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001414 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001415 nd->flags |= LOOKUP_JUMPED;
1416 }
Al Virofe479a52011-02-22 15:10:03 -05001417 break;
1418 case 1:
1419 type = LAST_DOT;
1420 }
Al Viro5a202bc2011-03-08 14:17:44 -05001421 if (likely(type == LAST_NORM)) {
1422 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001423 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001424 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1425 err = parent->d_op->d_hash(parent, nd->inode,
1426 &this);
1427 if (err < 0)
1428 break;
1429 }
1430 }
Al Virofe479a52011-02-22 15:10:03 -05001431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 /* remove trailing slashes? */
1433 if (!c)
1434 goto last_component;
1435 while (*++name == '/');
1436 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001437 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Al Viroce57dfc2011-03-13 19:58:58 -04001439 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1440 if (err < 0)
1441 return err;
Al Virofe479a52011-02-22 15:10:03 -05001442
Al Viroce57dfc2011-03-13 19:58:58 -04001443 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001444 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001446 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001447 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001448 if (can_lookup(nd->inode))
1449 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001451 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 /* here ends the main loop */
1453
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454last_component:
Al Virob3563792011-03-14 21:54:55 -04001455 nd->last = this;
1456 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001457 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
Al Viro951361f2011-03-04 14:44:37 -05001459 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 return err;
1461}
1462
Al Viro70e9b352011-03-05 21:12:22 -05001463static int path_init(int dfd, const char *name, unsigned int flags,
1464 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001466 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001467 int fput_needed;
1468 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001471 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001473 if (flags & LOOKUP_ROOT) {
1474 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001475 if (*name) {
1476 if (!inode->i_op->lookup)
1477 return -ENOTDIR;
1478 retval = inode_permission(inode, MAY_EXEC);
1479 if (retval)
1480 return retval;
1481 }
Al Viro5b6ca022011-03-09 23:04:47 -05001482 nd->path = nd->root;
1483 nd->inode = inode;
1484 if (flags & LOOKUP_RCU) {
1485 br_read_lock(vfsmount_lock);
1486 rcu_read_lock();
1487 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1488 } else {
1489 path_get(&nd->path);
1490 }
1491 return 0;
1492 }
1493
Al Viro2a737872009-04-07 11:49:53 -04001494 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001497 if (flags & LOOKUP_RCU) {
1498 br_read_lock(vfsmount_lock);
1499 rcu_read_lock();
1500 set_root_rcu(nd);
1501 } else {
1502 set_root(nd);
1503 path_get(&nd->root);
1504 }
Al Viro2a737872009-04-07 11:49:53 -04001505 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001506 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001507 if (flags & LOOKUP_RCU) {
1508 struct fs_struct *fs = current->fs;
1509 unsigned seq;
1510
1511 br_read_lock(vfsmount_lock);
1512 rcu_read_lock();
1513
1514 do {
1515 seq = read_seqcount_begin(&fs->seq);
1516 nd->path = fs->pwd;
1517 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1518 } while (read_seqcount_retry(&fs->seq, seq));
1519 } else {
1520 get_fs_pwd(current->fs, &nd->path);
1521 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001522 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001523 struct dentry *dentry;
1524
Al Viro1abf0c72011-03-13 03:51:11 -04001525 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001526 retval = -EBADF;
1527 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001528 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001529
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001530 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001531
Al Virof52e0c12011-03-14 18:56:51 -04001532 if (*name) {
1533 retval = -ENOTDIR;
1534 if (!S_ISDIR(dentry->d_inode->i_mode))
1535 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001536
Al Viro4ad5abb2011-06-20 19:57:03 -04001537 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001538 if (retval)
1539 goto fput_fail;
1540 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001541
Jan Blunck5dd784d02008-02-14 19:34:38 -08001542 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001543 if (flags & LOOKUP_RCU) {
1544 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001545 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001546 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1547 br_read_lock(vfsmount_lock);
1548 rcu_read_lock();
1549 } else {
1550 path_get(&file->f_path);
1551 fput_light(file, fput_needed);
1552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 }
Al Viroe41f7d42011-02-22 14:02:58 -05001554
Nick Piggin31e6b012011-01-07 17:49:52 +11001555 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001556 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001557
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001558fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001559 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001560out_fail:
1561 return retval;
1562}
1563
Al Virobd92d7f2011-03-14 19:54:59 -04001564static inline int lookup_last(struct nameidata *nd, struct path *path)
1565{
1566 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1567 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1568
1569 nd->flags &= ~LOOKUP_PARENT;
1570 return walk_component(nd, path, &nd->last, nd->last_type,
1571 nd->flags & LOOKUP_FOLLOW);
1572}
1573
Al Viro9b4a9b12009-04-07 11:44:16 -04001574/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001575static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001576 unsigned int flags, struct nameidata *nd)
1577{
Al Viro70e9b352011-03-05 21:12:22 -05001578 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001579 struct path path;
1580 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001581
1582 /*
1583 * Path walking is largely split up into 2 different synchronisation
1584 * schemes, rcu-walk and ref-walk (explained in
1585 * Documentation/filesystems/path-lookup.txt). These share much of the
1586 * path walk code, but some things particularly setup, cleanup, and
1587 * following mounts are sufficiently divergent that functions are
1588 * duplicated. Typically there is a function foo(), and its RCU
1589 * analogue, foo_rcu().
1590 *
1591 * -ECHILD is the error number of choice (just to avoid clashes) that
1592 * is returned if some aspect of an rcu-walk fails. Such an error must
1593 * be handled by restarting a traditional ref-walk (which will always
1594 * be able to complete).
1595 */
Al Virobd92d7f2011-03-14 19:54:59 -04001596 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001597
Al Virobd92d7f2011-03-14 19:54:59 -04001598 if (unlikely(err))
1599 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001600
1601 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001602 err = link_path_walk(name, nd);
1603
1604 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001605 err = lookup_last(nd, &path);
1606 while (err > 0) {
1607 void *cookie;
1608 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001609 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001610 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001611 if (!err)
1612 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001613 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001614 }
1615 }
Al Viroee0827c2011-02-21 23:38:09 -05001616
Al Viro9f1fafe2011-03-25 11:00:12 -04001617 if (!err)
1618 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001619
1620 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1621 if (!nd->inode->i_op->lookup) {
1622 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001623 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001624 }
1625 }
Al Viro16c2cd72011-02-22 15:50:10 -05001626
Al Viro70e9b352011-03-05 21:12:22 -05001627 if (base)
1628 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001629
Al Viro5b6ca022011-03-09 23:04:47 -05001630 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001631 path_put(&nd->root);
1632 nd->root.mnt = NULL;
1633 }
Al Virobd92d7f2011-03-14 19:54:59 -04001634 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001635}
Nick Piggin31e6b012011-01-07 17:49:52 +11001636
Al Viroee0827c2011-02-21 23:38:09 -05001637static int do_path_lookup(int dfd, const char *name,
1638 unsigned int flags, struct nameidata *nd)
1639{
1640 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1641 if (unlikely(retval == -ECHILD))
1642 retval = path_lookupat(dfd, name, flags, nd);
1643 if (unlikely(retval == -ESTALE))
1644 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001645
1646 if (likely(!retval)) {
1647 if (unlikely(!audit_dummy_context())) {
1648 if (nd->path.dentry && nd->inode)
1649 audit_inode(name, nd->path.dentry);
1650 }
1651 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001652 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653}
1654
Al Viroc9c6cac2011-02-16 15:15:47 -05001655int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001656{
Al Viroc9c6cac2011-02-16 15:15:47 -05001657 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001658}
1659
Al Virod1811462008-08-02 00:49:18 -04001660int kern_path(const char *name, unsigned int flags, struct path *path)
1661{
1662 struct nameidata nd;
1663 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1664 if (!res)
1665 *path = nd.path;
1666 return res;
1667}
1668
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001669/**
1670 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1671 * @dentry: pointer to dentry of the base directory
1672 * @mnt: pointer to vfs mount of the base directory
1673 * @name: pointer to file name
1674 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001675 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001676 */
1677int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1678 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001679 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001680{
Al Viroe0a01242011-06-27 17:00:37 -04001681 struct nameidata nd;
1682 int err;
1683 nd.root.dentry = dentry;
1684 nd.root.mnt = mnt;
1685 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001686 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001687 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1688 if (!err)
1689 *path = nd.path;
1690 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001691}
1692
Christoph Hellwigeead1912007-10-16 23:25:38 -07001693static struct dentry *__lookup_hash(struct qstr *name,
1694 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001696 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001697 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 int err;
1699
Al Viro4ad5abb2011-06-20 19:57:03 -04001700 err = inode_permission(inode, MAY_EXEC);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001701 if (err)
1702 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001705 * Don't bother with __d_lookup: callers are for creat as
1706 * well as unlink, so a lot of the time it would cost
1707 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001708 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001709 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001710
Josef Bacik44396f42011-05-31 11:58:49 -04001711 if (dentry && d_need_lookup(dentry)) {
1712 /*
1713 * __lookup_hash is called with the parent dir's i_mutex already
1714 * held, so we are good to go here.
1715 */
1716 dentry = d_inode_lookup(base, dentry, nd);
1717 if (IS_ERR(dentry))
1718 return dentry;
1719 }
1720
Al Virod2d9e9f2011-06-20 10:55:26 -04001721 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1722 int status = d_revalidate(dentry, nd);
1723 if (unlikely(status <= 0)) {
1724 /*
1725 * The dentry failed validation.
1726 * If d_revalidate returned 0 attempt to invalidate
1727 * the dentry otherwise d_revalidate is asking us
1728 * to return a fail status.
1729 */
1730 if (status < 0) {
1731 dput(dentry);
1732 return ERR_PTR(status);
1733 } else if (!d_invalidate(dentry)) {
1734 dput(dentry);
1735 dentry = NULL;
1736 }
1737 }
1738 }
Al Viro6e6b1bd2009-08-13 23:38:37 +04001739
Nick Pigginbaa03892010-08-18 04:37:31 +10001740 if (!dentry)
1741 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 return dentry;
1744}
1745
James Morris057f6c02007-04-26 00:12:05 -07001746/*
1747 * Restricted form of lookup. Doesn't follow links, single-component only,
1748 * needs parent already locked. Doesn't follow mounts.
1749 * SMP-safe.
1750 */
Adrian Bunka244e162006-03-31 02:32:11 -08001751static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
Jan Blunck4ac91372008-02-14 19:34:32 -08001753 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754}
1755
Christoph Hellwigeead1912007-10-16 23:25:38 -07001756/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001757 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001758 * @name: pathname component to lookup
1759 * @base: base directory to lookup from
1760 * @len: maximum length @len should be interpreted to
1761 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001762 * Note that this routine is purely a helper for filesystem usage and should
1763 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001764 * nameidata argument is passed to the filesystem methods and a filesystem
1765 * using this helper needs to be prepared for that.
1766 */
James Morris057f6c02007-04-26 00:12:05 -07001767struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1768{
James Morris057f6c02007-04-26 00:12:05 -07001769 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001770 unsigned long hash;
1771 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001772
David Woodhouse2f9092e2009-04-20 23:18:37 +01001773 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1774
Al Viro6a96ba52011-03-07 23:49:20 -05001775 this.name = name;
1776 this.len = len;
1777 if (!len)
1778 return ERR_PTR(-EACCES);
1779
1780 hash = init_name_hash();
1781 while (len--) {
1782 c = *(const unsigned char *)name++;
1783 if (c == '/' || c == '\0')
1784 return ERR_PTR(-EACCES);
1785 hash = partial_name_hash(c, hash);
1786 }
1787 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001788 /*
1789 * See if the low-level filesystem might want
1790 * to use its own hash..
1791 */
1792 if (base->d_flags & DCACHE_OP_HASH) {
1793 int err = base->d_op->d_hash(base, base->d_inode, &this);
1794 if (err < 0)
1795 return ERR_PTR(err);
1796 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001797
Christoph Hellwig49705b72005-11-08 21:35:06 -08001798 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001799}
1800
Al Viro2d8f3032008-07-22 09:59:21 -04001801int user_path_at(int dfd, const char __user *name, unsigned flags,
1802 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
Al Viro2d8f3032008-07-22 09:59:21 -04001804 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001805 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001808
1809 BUG_ON(flags & LOOKUP_PARENT);
1810
1811 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001813 if (!err)
1814 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
1816 return err;
1817}
1818
Al Viro2ad94ae2008-07-21 09:32:51 -04001819static int user_path_parent(int dfd, const char __user *path,
1820 struct nameidata *nd, char **name)
1821{
1822 char *s = getname(path);
1823 int error;
1824
1825 if (IS_ERR(s))
1826 return PTR_ERR(s);
1827
1828 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1829 if (error)
1830 putname(s);
1831 else
1832 *name = s;
1833
1834 return error;
1835}
1836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837/*
1838 * It's inline, so penalty for filesystems that don't use sticky bit is
1839 * minimal.
1840 */
1841static inline int check_sticky(struct inode *dir, struct inode *inode)
1842{
David Howellsda9592e2008-11-14 10:39:05 +11001843 uid_t fsuid = current_fsuid();
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 if (!(dir->i_mode & S_ISVTX))
1846 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001847 if (current_user_ns() != inode_userns(inode))
1848 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001849 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001851 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001853
1854other_userns:
1855 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
1858/*
1859 * Check whether we can remove a link victim from directory dir, check
1860 * whether the type of victim is right.
1861 * 1. We can't do it if dir is read-only (done in permission())
1862 * 2. We should have write and exec permissions on dir
1863 * 3. We can't remove anything from append-only dir
1864 * 4. We can't do anything with immutable dir (done in permission())
1865 * 5. If the sticky bit on dir is set we should either
1866 * a. be owner of dir, or
1867 * b. be owner of victim, or
1868 * c. have CAP_FOWNER capability
1869 * 6. If the victim is append-only or immutable we can't do antyhing with
1870 * links pointing to it.
1871 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1872 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1873 * 9. We can't remove a root or mountpoint.
1874 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1875 * nfs_async_unlink().
1876 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001877static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
1879 int error;
1880
1881 if (!victim->d_inode)
1882 return -ENOENT;
1883
1884 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001885 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Al Virof419a2e2008-07-22 00:07:17 -04001887 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 if (error)
1889 return error;
1890 if (IS_APPEND(dir))
1891 return -EPERM;
1892 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001893 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 return -EPERM;
1895 if (isdir) {
1896 if (!S_ISDIR(victim->d_inode->i_mode))
1897 return -ENOTDIR;
1898 if (IS_ROOT(victim))
1899 return -EBUSY;
1900 } else if (S_ISDIR(victim->d_inode->i_mode))
1901 return -EISDIR;
1902 if (IS_DEADDIR(dir))
1903 return -ENOENT;
1904 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1905 return -EBUSY;
1906 return 0;
1907}
1908
1909/* Check whether we can create an object with dentry child in directory
1910 * dir.
1911 * 1. We can't do it if child already exists (open has special treatment for
1912 * this case, but since we are inlined it's OK)
1913 * 2. We can't do it if dir is read-only (done in permission())
1914 * 3. We should have write and exec permissions on dir
1915 * 4. We can't do it if dir is immutable (done in permission())
1916 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001917static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918{
1919 if (child->d_inode)
1920 return -EEXIST;
1921 if (IS_DEADDIR(dir))
1922 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001923 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924}
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926/*
1927 * p1 and p2 should be directories on the same fs.
1928 */
1929struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1930{
1931 struct dentry *p;
1932
1933 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001934 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return NULL;
1936 }
1937
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001938 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001940 p = d_ancestor(p2, p1);
1941 if (p) {
1942 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1944 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 }
1946
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001947 p = d_ancestor(p1, p2);
1948 if (p) {
1949 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1950 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1951 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 }
1953
Ingo Molnarf2eace22006-07-03 00:25:05 -07001954 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1955 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 return NULL;
1957}
1958
1959void unlock_rename(struct dentry *p1, struct dentry *p2)
1960{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001961 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001963 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001964 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 }
1966}
1967
1968int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1969 struct nameidata *nd)
1970{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001971 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
1973 if (error)
1974 return error;
1975
Al Viroacfa4382008-12-04 10:06:33 -05001976 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 return -EACCES; /* shouldn't it be ENOSYS? */
1978 mode &= S_IALLUGO;
1979 mode |= S_IFREG;
1980 error = security_inode_create(dir, dentry, mode);
1981 if (error)
1982 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001984 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001985 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 return error;
1987}
1988
Al Viro73d049a2011-03-11 12:08:24 -05001989static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001991 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 struct inode *inode = dentry->d_inode;
1993 int error;
1994
Al Virobcda7652011-03-13 16:42:14 -04001995 /* O_PATH? */
1996 if (!acc_mode)
1997 return 0;
1998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 if (!inode)
2000 return -ENOENT;
2001
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002002 switch (inode->i_mode & S_IFMT) {
2003 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002005 case S_IFDIR:
2006 if (acc_mode & MAY_WRITE)
2007 return -EISDIR;
2008 break;
2009 case S_IFBLK:
2010 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002011 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002013 /*FALLTHRU*/
2014 case S_IFIFO:
2015 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002017 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002018 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002019
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002020 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002021 if (error)
2022 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002023
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 /*
2025 * An append-only file must be opened in append mode for writing.
2026 */
2027 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002028 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002029 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002031 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 }
2033
2034 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002035 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002036 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038 /*
2039 * Ensure there are no outstanding leases on the file.
2040 */
Al Virob65a9cf2009-12-16 06:27:40 -05002041 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002042}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Jeff Laytone1181ee2010-12-07 16:19:50 -05002044static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002045{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002046 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002047 struct inode *inode = path->dentry->d_inode;
2048 int error = get_write_access(inode);
2049 if (error)
2050 return error;
2051 /*
2052 * Refuse to truncate files with mandatory locks held on them.
2053 */
2054 error = locks_verify_locked(inode);
2055 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002056 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002057 if (!error) {
2058 error = do_truncate(path->dentry, 0,
2059 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002060 filp);
Al Viro7715b522009-12-16 03:54:00 -05002061 }
2062 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002063 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064}
2065
Dave Hansend57999e2008-02-15 14:37:27 -08002066static inline int open_to_namei_flags(int flag)
2067{
Al Viro8a5e9292011-06-25 19:15:54 -04002068 if ((flag & O_ACCMODE) == 3)
2069 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002070 return flag;
2071}
2072
Nick Piggin31e6b012011-01-07 17:49:52 +11002073/*
Al Virofe2d35f2011-03-05 22:58:25 -05002074 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002075 */
Al Virofb1cc552009-12-24 01:58:28 -05002076static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002077 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002078{
Al Viroa1e28032009-12-24 02:12:06 -05002079 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002080 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002081 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002082 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002083 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002084 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002085 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002086 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002087
Al Viroc3e380b2011-02-23 13:39:45 -05002088 nd->flags &= ~LOOKUP_PARENT;
2089 nd->flags |= op->intent;
2090
Al Viro1f36f772009-12-26 10:56:19 -05002091 switch (nd->last_type) {
2092 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002093 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002094 error = handle_dots(nd, nd->last_type);
2095 if (error)
2096 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002097 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002098 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002099 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002100 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002101 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002102 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002103 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002104 error = -EISDIR;
2105 goto exit;
2106 }
2107 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002108 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002109 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002110 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002111 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002112 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002113 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002114 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002115
Al Viroca344a892011-03-09 00:36:45 -05002116 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002117 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002118 if (nd->last.name[nd->last.len])
2119 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002120 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2121 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002122 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002123 error = walk_component(nd, path, &nd->last, LAST_NORM,
2124 !symlink_ok);
2125 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002126 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002127 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002128 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002129 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002130 error = complete_walk(nd);
2131 if (error)
2132 return ERR_PTR(-ECHILD);
Al Virofe2d35f2011-03-05 22:58:25 -05002133
2134 error = -ENOTDIR;
2135 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002136 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002137 goto exit;
2138 }
2139 audit_inode(pathname, nd->path.dentry);
2140 goto ok;
2141 }
2142
2143 /* create side of things */
Al Viro9f1fafe2011-03-25 11:00:12 -04002144 error = complete_walk(nd);
2145 if (error)
2146 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002147
2148 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002149 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002150 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002151 if (nd->last.name[nd->last.len])
2152 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002153
Al Viroa1e28032009-12-24 02:12:06 -05002154 mutex_lock(&dir->d_inode->i_mutex);
2155
Al Viro6c0d46c2011-03-09 00:59:59 -05002156 dentry = lookup_hash(nd);
2157 error = PTR_ERR(dentry);
2158 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002159 mutex_unlock(&dir->d_inode->i_mutex);
2160 goto exit;
2161 }
2162
Al Viro6c0d46c2011-03-09 00:59:59 -05002163 path->dentry = dentry;
2164 path->mnt = nd->path.mnt;
2165
Al Virofb1cc552009-12-24 01:58:28 -05002166 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002167 if (!dentry->d_inode) {
2168 int mode = op->mode;
2169 if (!IS_POSIXACL(dir->d_inode))
2170 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002171 /*
2172 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002173 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002174 * the time when the file is created and when
2175 * a permanent write count is taken through
2176 * the 'struct file' in nameidata_to_filp().
2177 */
2178 error = mnt_want_write(nd->path.mnt);
2179 if (error)
2180 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002181 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002182 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002183 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002184 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002185 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002186 error = security_path_mknod(&nd->path, dentry, mode, 0);
2187 if (error)
2188 goto exit_mutex_unlock;
2189 error = vfs_create(dir->d_inode, dentry, mode, nd);
2190 if (error)
2191 goto exit_mutex_unlock;
2192 mutex_unlock(&dir->d_inode->i_mutex);
2193 dput(nd->path.dentry);
2194 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002195 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002196 }
2197
2198 /*
2199 * It already exists.
2200 */
2201 mutex_unlock(&dir->d_inode->i_mutex);
2202 audit_inode(pathname, path->dentry);
2203
2204 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002205 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002206 goto exit_dput;
2207
David Howells9875cf82011-01-14 18:45:21 +00002208 error = follow_managed(path, nd->flags);
2209 if (error < 0)
2210 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002211
2212 error = -ENOENT;
2213 if (!path->dentry->d_inode)
2214 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002215
2216 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002217 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002218
2219 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002220 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002221 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002222 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002223 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002224ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002225 if (!S_ISREG(nd->inode->i_mode))
2226 will_truncate = 0;
2227
Al Viro0f9d1a12011-03-09 00:13:14 -05002228 if (will_truncate) {
2229 error = mnt_want_write(nd->path.mnt);
2230 if (error)
2231 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002232 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002233 }
Al Viroca344a892011-03-09 00:36:45 -05002234common:
Al Virobcda7652011-03-13 16:42:14 -04002235 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002236 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002237 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002238 filp = nameidata_to_filp(nd);
2239 if (!IS_ERR(filp)) {
2240 error = ima_file_check(filp, op->acc_mode);
2241 if (error) {
2242 fput(filp);
2243 filp = ERR_PTR(error);
2244 }
2245 }
2246 if (!IS_ERR(filp)) {
2247 if (will_truncate) {
2248 error = handle_truncate(filp);
2249 if (error) {
2250 fput(filp);
2251 filp = ERR_PTR(error);
2252 }
2253 }
2254 }
Al Viroca344a892011-03-09 00:36:45 -05002255out:
2256 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002257 mnt_drop_write(nd->path.mnt);
2258 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002259 return filp;
2260
2261exit_mutex_unlock:
2262 mutex_unlock(&dir->d_inode->i_mutex);
2263exit_dput:
2264 path_put_conditional(path, nd);
2265exit:
Al Viroca344a892011-03-09 00:36:45 -05002266 filp = ERR_PTR(error);
2267 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002268}
2269
Al Viro13aab422011-02-23 17:54:08 -05002270static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002271 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272{
Al Virofe2d35f2011-03-05 22:58:25 -05002273 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002274 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002275 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002276 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002277
2278 filp = get_empty_filp();
2279 if (!filp)
2280 return ERR_PTR(-ENFILE);
2281
Al Viro47c805d2011-02-23 17:44:09 -05002282 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002283 nd->intent.open.file = filp;
2284 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2285 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002286
Al Viro73d049a2011-03-11 12:08:24 -05002287 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002288 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002289 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002290
Al Virofe2d35f2011-03-05 22:58:25 -05002291 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002292 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002293 if (unlikely(error))
2294 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Al Viro73d049a2011-03-11 12:08:24 -05002296 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002297 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002298 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002299 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002300 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002301 path_put_conditional(&path, nd);
2302 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002303 filp = ERR_PTR(-ELOOP);
2304 break;
2305 }
Al Viro73d049a2011-03-11 12:08:24 -05002306 nd->flags |= LOOKUP_PARENT;
2307 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002308 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002309 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002310 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002311 else
Al Viro73d049a2011-03-11 12:08:24 -05002312 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002313 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002314 }
Al Viro10fa8e62009-12-26 07:09:49 -05002315out:
Al Viro73d049a2011-03-11 12:08:24 -05002316 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2317 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002318 if (base)
2319 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002320 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002321 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Nick Piggin31e6b012011-01-07 17:49:52 +11002323out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002324 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002325 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326}
2327
Al Viro13aab422011-02-23 17:54:08 -05002328struct file *do_filp_open(int dfd, const char *pathname,
2329 const struct open_flags *op, int flags)
2330{
Al Viro73d049a2011-03-11 12:08:24 -05002331 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002332 struct file *filp;
2333
Al Viro73d049a2011-03-11 12:08:24 -05002334 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002335 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002336 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002337 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002338 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002339 return filp;
2340}
2341
Al Viro73d049a2011-03-11 12:08:24 -05002342struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2343 const char *name, const struct open_flags *op, int flags)
2344{
2345 struct nameidata nd;
2346 struct file *file;
2347
2348 nd.root.mnt = mnt;
2349 nd.root.dentry = dentry;
2350
2351 flags |= LOOKUP_ROOT;
2352
Al Virobcda7652011-03-13 16:42:14 -04002353 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002354 return ERR_PTR(-ELOOP);
2355
2356 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2357 if (unlikely(file == ERR_PTR(-ECHILD)))
2358 file = path_openat(-1, name, &nd, op, flags);
2359 if (unlikely(file == ERR_PTR(-ESTALE)))
2360 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2361 return file;
2362}
2363
Al Viroed75e952011-06-27 16:53:43 -04002364struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002366 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002367 struct nameidata nd;
2368 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2369 if (error)
2370 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002372 /*
2373 * Yucky last component or no last component at all?
2374 * (foo/., foo/.., /////)
2375 */
Al Viroed75e952011-06-27 16:53:43 -04002376 if (nd.last_type != LAST_NORM)
2377 goto out;
2378 nd.flags &= ~LOOKUP_PARENT;
2379 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2380 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002381
2382 /*
2383 * Do the final lookup.
2384 */
Al Viroed75e952011-06-27 16:53:43 -04002385 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2386 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 if (IS_ERR(dentry))
2388 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002389
Al Viroe9baf6e2008-05-15 04:49:12 -04002390 if (dentry->d_inode)
2391 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002392 /*
2393 * Special case - lookup gave negative, but... we had foo/bar/
2394 * From the vfs_mknod() POV we just have a negative dentry -
2395 * all is fine. Let's be bastards - you had / on the end, you've
2396 * been asking for (non-existent) directory. -ENOENT for you.
2397 */
Al Viroed75e952011-06-27 16:53:43 -04002398 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002399 dput(dentry);
2400 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002401 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002402 }
Al Viroed75e952011-06-27 16:53:43 -04002403 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002405eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002407 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408fail:
Al Viroed75e952011-06-27 16:53:43 -04002409 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2410out:
2411 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 return dentry;
2413}
Al Virodae6ad82011-06-26 11:50:15 -04002414EXPORT_SYMBOL(kern_path_create);
2415
2416struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2417{
2418 char *tmp = getname(pathname);
2419 struct dentry *res;
2420 if (IS_ERR(tmp))
2421 return ERR_CAST(tmp);
2422 res = kern_path_create(dfd, tmp, path, is_dir);
2423 putname(tmp);
2424 return res;
2425}
2426EXPORT_SYMBOL(user_path_create);
2427
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2429{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002430 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
2432 if (error)
2433 return error;
2434
Serge E. Hallyne795b712011-03-23 16:43:25 -07002435 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2436 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 return -EPERM;
2438
Al Viroacfa4382008-12-04 10:06:33 -05002439 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 return -EPERM;
2441
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002442 error = devcgroup_inode_mknod(mode, dev);
2443 if (error)
2444 return error;
2445
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 error = security_inode_mknod(dir, dentry, mode, dev);
2447 if (error)
2448 return error;
2449
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002451 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002452 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 return error;
2454}
2455
Dave Hansen463c3192008-02-15 14:37:57 -08002456static int may_mknod(mode_t mode)
2457{
2458 switch (mode & S_IFMT) {
2459 case S_IFREG:
2460 case S_IFCHR:
2461 case S_IFBLK:
2462 case S_IFIFO:
2463 case S_IFSOCK:
2464 case 0: /* zero mode translates to S_IFREG */
2465 return 0;
2466 case S_IFDIR:
2467 return -EPERM;
2468 default:
2469 return -EINVAL;
2470 }
2471}
2472
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002473SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2474 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475{
Al Viro2ad94ae2008-07-21 09:32:51 -04002476 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002477 struct path path;
2478 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
2480 if (S_ISDIR(mode))
2481 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
Al Virodae6ad82011-06-26 11:50:15 -04002483 dentry = user_path_create(dfd, filename, &path, 0);
2484 if (IS_ERR(dentry))
2485 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002486
Al Virodae6ad82011-06-26 11:50:15 -04002487 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002488 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002489 error = may_mknod(mode);
2490 if (error)
2491 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002492 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002493 if (error)
2494 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002495 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002496 if (error)
2497 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002498 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002500 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 break;
2502 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002503 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 new_decode_dev(dev));
2505 break;
2506 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002507 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002510out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002511 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002512out_dput:
2513 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002514 mutex_unlock(&path.dentry->d_inode->i_mutex);
2515 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
2517 return error;
2518}
2519
Heiko Carstens3480b252009-01-14 14:14:16 +01002520SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002521{
2522 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2523}
2524
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2526{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002527 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
2529 if (error)
2530 return error;
2531
Al Viroacfa4382008-12-04 10:06:33 -05002532 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 return -EPERM;
2534
2535 mode &= (S_IRWXUGO|S_ISVTX);
2536 error = security_inode_mkdir(dir, dentry, mode);
2537 if (error)
2538 return error;
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002541 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002542 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 return error;
2544}
2545
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002546SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547{
Dave Hansen6902d922006-09-30 23:29:01 -07002548 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002549 struct path path;
2550 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551
Al Virodae6ad82011-06-26 11:50:15 -04002552 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002553 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002554 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002555
Al Virodae6ad82011-06-26 11:50:15 -04002556 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002557 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002558 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002559 if (error)
2560 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002561 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002562 if (error)
2563 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002564 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002565out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002566 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002567out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002568 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002569 mutex_unlock(&path.dentry->d_inode->i_mutex);
2570 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 return error;
2572}
2573
Heiko Carstens3cdad422009-01-14 14:14:22 +01002574SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002575{
2576 return sys_mkdirat(AT_FDCWD, pathname, mode);
2577}
2578
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579/*
Sage Weila71905f2011-05-24 13:06:08 -07002580 * The dentry_unhash() helper will try to drop the dentry early: we
2581 * should have a usage count of 2 if we're the only user of this
2582 * dentry, and if that is true (possibly after pruning the dcache),
2583 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 *
2585 * A low-level filesystem can, if it choses, legally
2586 * do a
2587 *
2588 * if (!d_unhashed(dentry))
2589 * return -EBUSY;
2590 *
2591 * if it cannot handle the case of removing a directory
2592 * that is still in use by something else..
2593 */
2594void dentry_unhash(struct dentry *dentry)
2595{
Vasily Averindc168422006-12-06 20:37:07 -08002596 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002598 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 __d_drop(dentry);
2600 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601}
2602
2603int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2604{
2605 int error = may_delete(dir, dentry, 1);
2606
2607 if (error)
2608 return error;
2609
Al Viroacfa4382008-12-04 10:06:33 -05002610 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 return -EPERM;
2612
Al Viro1d2ef592011-09-14 18:55:41 +01002613 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002614 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
Sage Weil912dbc12011-05-24 13:06:11 -07002616 error = -EBUSY;
2617 if (d_mountpoint(dentry))
2618 goto out;
2619
2620 error = security_inode_rmdir(dir, dentry);
2621 if (error)
2622 goto out;
2623
Sage Weil3cebde22011-05-29 21:20:59 -07002624 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002625 error = dir->i_op->rmdir(dir, dentry);
2626 if (error)
2627 goto out;
2628
2629 dentry->d_inode->i_flags |= S_DEAD;
2630 dont_mount(dentry);
2631
2632out:
2633 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01002634 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002635 if (!error)
2636 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 return error;
2638}
2639
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002640static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641{
2642 int error = 0;
2643 char * name;
2644 struct dentry *dentry;
2645 struct nameidata nd;
2646
Al Viro2ad94ae2008-07-21 09:32:51 -04002647 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002649 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
2651 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002652 case LAST_DOTDOT:
2653 error = -ENOTEMPTY;
2654 goto exit1;
2655 case LAST_DOT:
2656 error = -EINVAL;
2657 goto exit1;
2658 case LAST_ROOT:
2659 error = -EBUSY;
2660 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002662
2663 nd.flags &= ~LOOKUP_PARENT;
2664
Jan Blunck4ac91372008-02-14 19:34:32 -08002665 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002666 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002668 if (IS_ERR(dentry))
2669 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002670 if (!dentry->d_inode) {
2671 error = -ENOENT;
2672 goto exit3;
2673 }
Dave Hansen06227532008-02-15 14:37:34 -08002674 error = mnt_want_write(nd.path.mnt);
2675 if (error)
2676 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002677 error = security_path_rmdir(&nd.path, dentry);
2678 if (error)
2679 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002680 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002681exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002682 mnt_drop_write(nd.path.mnt);
2683exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002684 dput(dentry);
2685exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002686 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002688 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 putname(name);
2690 return error;
2691}
2692
Heiko Carstens3cdad422009-01-14 14:14:22 +01002693SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002694{
2695 return do_rmdir(AT_FDCWD, pathname);
2696}
2697
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698int vfs_unlink(struct inode *dir, struct dentry *dentry)
2699{
2700 int error = may_delete(dir, dentry, 0);
2701
2702 if (error)
2703 return error;
2704
Al Viroacfa4382008-12-04 10:06:33 -05002705 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 return -EPERM;
2707
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002708 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 if (d_mountpoint(dentry))
2710 error = -EBUSY;
2711 else {
2712 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002713 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002715 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002716 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002719 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
2721 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2722 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002723 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 }
Robert Love0eeca282005-07-12 17:06:03 -04002726
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 return error;
2728}
2729
2730/*
2731 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002732 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 * writeout happening, and we don't want to prevent access to the directory
2734 * while waiting on the I/O.
2735 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002736static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
Al Viro2ad94ae2008-07-21 09:32:51 -04002738 int error;
2739 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 struct dentry *dentry;
2741 struct nameidata nd;
2742 struct inode *inode = NULL;
2743
Al Viro2ad94ae2008-07-21 09:32:51 -04002744 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002746 return error;
2747
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 error = -EISDIR;
2749 if (nd.last_type != LAST_NORM)
2750 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002751
2752 nd.flags &= ~LOOKUP_PARENT;
2753
Jan Blunck4ac91372008-02-14 19:34:32 -08002754 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002755 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 error = PTR_ERR(dentry);
2757 if (!IS_ERR(dentry)) {
2758 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002759 if (nd.last.name[nd.last.len])
2760 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002762 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002763 goto slashes;
2764 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002765 error = mnt_want_write(nd.path.mnt);
2766 if (error)
2767 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002768 error = security_path_unlink(&nd.path, dentry);
2769 if (error)
2770 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002771 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002772exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002773 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 exit2:
2775 dput(dentry);
2776 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002777 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 if (inode)
2779 iput(inode); /* truncate the inode here */
2780exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002781 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 putname(name);
2783 return error;
2784
2785slashes:
2786 error = !dentry->d_inode ? -ENOENT :
2787 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2788 goto exit2;
2789}
2790
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002791SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002792{
2793 if ((flag & ~AT_REMOVEDIR) != 0)
2794 return -EINVAL;
2795
2796 if (flag & AT_REMOVEDIR)
2797 return do_rmdir(dfd, pathname);
2798
2799 return do_unlinkat(dfd, pathname);
2800}
2801
Heiko Carstens3480b252009-01-14 14:14:16 +01002802SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002803{
2804 return do_unlinkat(AT_FDCWD, pathname);
2805}
2806
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002807int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002809 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
2811 if (error)
2812 return error;
2813
Al Viroacfa4382008-12-04 10:06:33 -05002814 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 return -EPERM;
2816
2817 error = security_inode_symlink(dir, dentry, oldname);
2818 if (error)
2819 return error;
2820
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002822 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002823 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 return error;
2825}
2826
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002827SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2828 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829{
Al Viro2ad94ae2008-07-21 09:32:51 -04002830 int error;
2831 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07002832 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002833 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834
2835 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002836 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002838
Al Virodae6ad82011-06-26 11:50:15 -04002839 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07002840 error = PTR_ERR(dentry);
2841 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002842 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07002843
Al Virodae6ad82011-06-26 11:50:15 -04002844 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002845 if (error)
2846 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002847 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002848 if (error)
2849 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002850 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002851out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002852 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002853out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002854 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002855 mutex_unlock(&path.dentry->d_inode->i_mutex);
2856 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07002857out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 putname(from);
2859 return error;
2860}
2861
Heiko Carstens3480b252009-01-14 14:14:16 +01002862SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002863{
2864 return sys_symlinkat(oldname, AT_FDCWD, newname);
2865}
2866
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2868{
2869 struct inode *inode = old_dentry->d_inode;
2870 int error;
2871
2872 if (!inode)
2873 return -ENOENT;
2874
Miklos Szeredia95164d2008-07-30 15:08:48 +02002875 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 if (error)
2877 return error;
2878
2879 if (dir->i_sb != inode->i_sb)
2880 return -EXDEV;
2881
2882 /*
2883 * A link to an append-only or immutable file cannot be created.
2884 */
2885 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2886 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002887 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002889 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 return -EPERM;
2891
2892 error = security_inode_link(old_dentry, dir, new_dentry);
2893 if (error)
2894 return error;
2895
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002896 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302897 /* Make sure we don't allow creating hardlink to an unlinked file */
2898 if (inode->i_nlink == 0)
2899 error = -ENOENT;
2900 else
2901 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002902 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002903 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002904 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 return error;
2906}
2907
2908/*
2909 * Hardlinks are often used in delicate situations. We avoid
2910 * security-related surprises by not following symlinks on the
2911 * newname. --KAB
2912 *
2913 * We don't follow them on the oldname either to be compatible
2914 * with linux 2.0, and to avoid hard-linking to directories
2915 * and other special files. --ADM
2916 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002917SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2918 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919{
2920 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002921 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302922 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302925 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002926 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302927 /*
2928 * To use null names we require CAP_DAC_READ_SEARCH
2929 * This ensures that not everyone will be able to create
2930 * handlink using the passed filedescriptor.
2931 */
2932 if (flags & AT_EMPTY_PATH) {
2933 if (!capable(CAP_DAC_READ_SEARCH))
2934 return -ENOENT;
2935 how = LOOKUP_EMPTY;
2936 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002937
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302938 if (flags & AT_SYMLINK_FOLLOW)
2939 how |= LOOKUP_FOLLOW;
2940
2941 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002943 return error;
2944
Al Virodae6ad82011-06-26 11:50:15 -04002945 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002947 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002948 goto out;
2949
2950 error = -EXDEV;
2951 if (old_path.mnt != new_path.mnt)
2952 goto out_dput;
2953 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002954 if (error)
2955 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002956 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002957 if (error)
2958 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002959 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002960out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002961 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002962out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002963 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002964 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2965 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966out:
Al Viro2d8f3032008-07-22 09:59:21 -04002967 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968
2969 return error;
2970}
2971
Heiko Carstens3480b252009-01-14 14:14:16 +01002972SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002973{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002974 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002975}
2976
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977/*
2978 * The worst of all namespace operations - renaming directory. "Perverted"
2979 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2980 * Problems:
2981 * a) we can get into loop creation. Check is done in is_subdir().
2982 * b) race potential - two innocent renames can create a loop together.
2983 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002984 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 * story.
2986 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002987 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 * whether the target exists). Solution: try to be smart with locking
2989 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002990 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 * move will be locked. Thus we can rank directories by the tree
2992 * (ancestors first) and rank all non-directories after them.
2993 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002994 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 * HOWEVER, it relies on the assumption that any object with ->lookup()
2996 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2997 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07002998 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002999 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003001 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 * locking].
3003 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003004static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3005 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006{
3007 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003008 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
3010 /*
3011 * If we are going to change the parent - check write permissions,
3012 * we'll need to flip '..'.
3013 */
3014 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003015 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 if (error)
3017 return error;
3018 }
3019
3020 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3021 if (error)
3022 return error;
3023
Al Viro1d2ef592011-09-14 18:55:41 +01003024 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003025 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003026 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003027
3028 error = -EBUSY;
3029 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3030 goto out;
3031
Sage Weil3cebde22011-05-29 21:20:59 -07003032 if (target)
3033 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003034 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3035 if (error)
3036 goto out;
3037
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003039 target->i_flags |= S_DEAD;
3040 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 }
Sage Weil9055cba2011-05-24 13:06:12 -07003042out:
3043 if (target)
3044 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003045 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003046 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003047 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3048 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 return error;
3050}
3051
Adrian Bunk75c96f82005-05-05 16:16:09 -07003052static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3053 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054{
Sage Weil51892bb2011-05-24 13:06:13 -07003055 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 int error;
3057
3058 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3059 if (error)
3060 return error;
3061
3062 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003064 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003065
3066 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003068 goto out;
3069
3070 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3071 if (error)
3072 goto out;
3073
3074 if (target)
3075 dont_mount(new_dentry);
3076 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3077 d_move(old_dentry, new_dentry);
3078out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003080 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 dput(new_dentry);
3082 return error;
3083}
3084
3085int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3086 struct inode *new_dir, struct dentry *new_dentry)
3087{
3088 int error;
3089 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003090 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091
3092 if (old_dentry->d_inode == new_dentry->d_inode)
3093 return 0;
3094
3095 error = may_delete(old_dir, old_dentry, is_dir);
3096 if (error)
3097 return error;
3098
3099 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003100 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 else
3102 error = may_delete(new_dir, new_dentry, is_dir);
3103 if (error)
3104 return error;
3105
Al Viroacfa4382008-12-04 10:06:33 -05003106 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 return -EPERM;
3108
Robert Love0eeca282005-07-12 17:06:03 -04003109 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3110
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 if (is_dir)
3112 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3113 else
3114 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003115 if (!error)
3116 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003117 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003118 fsnotify_oldname_free(old_name);
3119
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 return error;
3121}
3122
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003123SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3124 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125{
Al Viro2ad94ae2008-07-21 09:32:51 -04003126 struct dentry *old_dir, *new_dir;
3127 struct dentry *old_dentry, *new_dentry;
3128 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003130 char *from;
3131 char *to;
3132 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133
Al Viro2ad94ae2008-07-21 09:32:51 -04003134 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 if (error)
3136 goto exit;
3137
Al Viro2ad94ae2008-07-21 09:32:51 -04003138 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 if (error)
3140 goto exit1;
3141
3142 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003143 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 goto exit2;
3145
Jan Blunck4ac91372008-02-14 19:34:32 -08003146 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 error = -EBUSY;
3148 if (oldnd.last_type != LAST_NORM)
3149 goto exit2;
3150
Jan Blunck4ac91372008-02-14 19:34:32 -08003151 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 if (newnd.last_type != LAST_NORM)
3153 goto exit2;
3154
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003155 oldnd.flags &= ~LOOKUP_PARENT;
3156 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003157 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003158
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 trap = lock_rename(new_dir, old_dir);
3160
Christoph Hellwig49705b72005-11-08 21:35:06 -08003161 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 error = PTR_ERR(old_dentry);
3163 if (IS_ERR(old_dentry))
3164 goto exit3;
3165 /* source must exist */
3166 error = -ENOENT;
3167 if (!old_dentry->d_inode)
3168 goto exit4;
3169 /* unless the source is a directory trailing slashes give -ENOTDIR */
3170 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3171 error = -ENOTDIR;
3172 if (oldnd.last.name[oldnd.last.len])
3173 goto exit4;
3174 if (newnd.last.name[newnd.last.len])
3175 goto exit4;
3176 }
3177 /* source should not be ancestor of target */
3178 error = -EINVAL;
3179 if (old_dentry == trap)
3180 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003181 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 error = PTR_ERR(new_dentry);
3183 if (IS_ERR(new_dentry))
3184 goto exit4;
3185 /* target should not be an ancestor of source */
3186 error = -ENOTEMPTY;
3187 if (new_dentry == trap)
3188 goto exit5;
3189
Dave Hansen9079b1e2008-02-15 14:37:49 -08003190 error = mnt_want_write(oldnd.path.mnt);
3191 if (error)
3192 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003193 error = security_path_rename(&oldnd.path, old_dentry,
3194 &newnd.path, new_dentry);
3195 if (error)
3196 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 error = vfs_rename(old_dir->d_inode, old_dentry,
3198 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003199exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003200 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201exit5:
3202 dput(new_dentry);
3203exit4:
3204 dput(old_dentry);
3205exit3:
3206 unlock_rename(new_dir, old_dir);
3207exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003208 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003209 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003211 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003213exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 return error;
3215}
3216
Heiko Carstensa26eab22009-01-14 14:14:17 +01003217SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003218{
3219 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3220}
3221
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3223{
3224 int len;
3225
3226 len = PTR_ERR(link);
3227 if (IS_ERR(link))
3228 goto out;
3229
3230 len = strlen(link);
3231 if (len > (unsigned) buflen)
3232 len = buflen;
3233 if (copy_to_user(buffer, link, len))
3234 len = -EFAULT;
3235out:
3236 return len;
3237}
3238
3239/*
3240 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3241 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3242 * using) it for any given inode is up to filesystem.
3243 */
3244int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3245{
3246 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003247 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003248 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003249
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003251 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003252 if (IS_ERR(cookie))
3253 return PTR_ERR(cookie);
3254
3255 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3256 if (dentry->d_inode->i_op->put_link)
3257 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3258 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259}
3260
3261int vfs_follow_link(struct nameidata *nd, const char *link)
3262{
3263 return __vfs_follow_link(nd, link);
3264}
3265
3266/* get the link contents into pagecache */
3267static char *page_getlink(struct dentry * dentry, struct page **ppage)
3268{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003269 char *kaddr;
3270 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003272 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003274 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003276 kaddr = kmap(page);
3277 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3278 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279}
3280
3281int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3282{
3283 struct page *page = NULL;
3284 char *s = page_getlink(dentry, &page);
3285 int res = vfs_readlink(dentry,buffer,buflen,s);
3286 if (page) {
3287 kunmap(page);
3288 page_cache_release(page);
3289 }
3290 return res;
3291}
3292
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003293void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003295 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003297 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298}
3299
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003300void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003302 struct page *page = cookie;
3303
3304 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305 kunmap(page);
3306 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 }
3308}
3309
Nick Piggin54566b22009-01-04 12:00:53 -08003310/*
3311 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3312 */
3313int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314{
3315 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003316 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003317 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003318 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003320 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3321 if (nofs)
3322 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323
NeilBrown7e53cac2006-03-25 03:07:57 -08003324retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003325 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003326 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003328 goto fail;
3329
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 kaddr = kmap_atomic(page, KM_USER0);
3331 memcpy(kaddr, symname, len-1);
3332 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003333
3334 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3335 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336 if (err < 0)
3337 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003338 if (err < len-1)
3339 goto retry;
3340
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 mark_inode_dirty(inode);
3342 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343fail:
3344 return err;
3345}
3346
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003347int page_symlink(struct inode *inode, const char *symname, int len)
3348{
3349 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003350 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003351}
3352
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003353const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 .readlink = generic_readlink,
3355 .follow_link = page_follow_link_light,
3356 .put_link = page_put_link,
3357};
3358
Al Viro2d8f3032008-07-22 09:59:21 -04003359EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003360EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361EXPORT_SYMBOL(follow_down);
3362EXPORT_SYMBOL(follow_up);
3363EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3364EXPORT_SYMBOL(getname);
3365EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366EXPORT_SYMBOL(lookup_one_len);
3367EXPORT_SYMBOL(page_follow_link_light);
3368EXPORT_SYMBOL(page_put_link);
3369EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003370EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371EXPORT_SYMBOL(page_symlink);
3372EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003373EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003374EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003375EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376EXPORT_SYMBOL(unlock_rename);
3377EXPORT_SYMBOL(vfs_create);
3378EXPORT_SYMBOL(vfs_follow_link);
3379EXPORT_SYMBOL(vfs_link);
3380EXPORT_SYMBOL(vfs_mkdir);
3381EXPORT_SYMBOL(vfs_mknod);
3382EXPORT_SYMBOL(generic_permission);
3383EXPORT_SYMBOL(vfs_readlink);
3384EXPORT_SYMBOL(vfs_rename);
3385EXPORT_SYMBOL(vfs_rmdir);
3386EXPORT_SYMBOL(vfs_symlink);
3387EXPORT_SYMBOL(vfs_unlink);
3388EXPORT_SYMBOL(dentry_unhash);
3389EXPORT_SYMBOL(generic_readlink);