blob: b52bc685465f5ecf1257d7982887b1435a7a6c2d [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 Blunck5dd784d2008-02-14 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
David Howells6f45b652011-01-14 18:45:31 +0000724 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
725 * and this is the terminal part of the path.
726 */
Al Viro49084c32011-06-25 21:59:52 -0400727 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
David Howells6f45b652011-01-14 18:45:31 +0000728 return -EISDIR; /* we actually want to stop here */
729
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200730 /* We don't want to mount if someone's just doing a stat -
731 * unless they're stat'ing a directory and appended a '/' to
732 * the name.
733 *
734 * We do, however, want to mount if someone wants to open or
735 * create a file of any type under the mountpoint, wants to
736 * traverse through the mountpoint or wants to open the
737 * mounted directory. Also, autofs may mark negative dentries
738 * as being automount points. These will need the attentions
739 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000740 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200741 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
742 LOOKUP_OPEN | LOOKUP_CREATE)) &&
743 path->dentry->d_inode)
744 return -EISDIR;
745
David Howells9875cf82011-01-14 18:45:21 +0000746 current->total_link_count++;
747 if (current->total_link_count >= 40)
748 return -ELOOP;
749
750 mnt = path->dentry->d_op->d_automount(path);
751 if (IS_ERR(mnt)) {
752 /*
753 * The filesystem is allowed to return -EISDIR here to indicate
754 * it doesn't want to automount. For instance, autofs would do
755 * this so that its userspace daemon can mount on this dentry.
756 *
757 * However, we can only permit this if it's a terminal point in
758 * the path being looked up; if it wasn't then the remainder of
759 * the path is inaccessible and we should say so.
760 */
Al Viro49084c32011-06-25 21:59:52 -0400761 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000762 return -EREMOTE;
763 return PTR_ERR(mnt);
764 }
David Howellsea5b7782011-01-14 19:10:03 +0000765
David Howells9875cf82011-01-14 18:45:21 +0000766 if (!mnt) /* mount collision */
767 return 0;
768
Al Viro8aef1882011-06-16 15:10:06 +0100769 if (!*need_mntput) {
770 /* lock_mount() may release path->mnt on error */
771 mntget(path->mnt);
772 *need_mntput = true;
773 }
Al Viro19a167a2011-01-17 01:35:23 -0500774 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000775
David Howellsea5b7782011-01-14 19:10:03 +0000776 switch (err) {
777 case -EBUSY:
778 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500779 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000780 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100781 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000782 path->mnt = mnt;
783 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000784 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500785 default:
786 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000787 }
Al Viro19a167a2011-01-17 01:35:23 -0500788
David Howells9875cf82011-01-14 18:45:21 +0000789}
790
791/*
792 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000793 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000794 * - Flagged as mountpoint
795 * - Flagged as automount point
796 *
797 * This may only be called in refwalk mode.
798 *
799 * Serialization is taken care of in namespace.c
800 */
801static int follow_managed(struct path *path, unsigned flags)
802{
Al Viro8aef1882011-06-16 15:10:06 +0100803 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000804 unsigned managed;
805 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100806 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000807
808 /* Given that we're not holding a lock here, we retain the value in a
809 * local variable for each dentry as we look at it so that we don't see
810 * the components of that value change under us */
811 while (managed = ACCESS_ONCE(path->dentry->d_flags),
812 managed &= DCACHE_MANAGED_DENTRY,
813 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000814 /* Allow the filesystem to manage the transit without i_mutex
815 * being held. */
816 if (managed & DCACHE_MANAGE_TRANSIT) {
817 BUG_ON(!path->dentry->d_op);
818 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400819 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000820 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100821 break;
David Howellscc53ce52011-01-14 18:45:26 +0000822 }
823
David Howells9875cf82011-01-14 18:45:21 +0000824 /* Transit to a mounted filesystem. */
825 if (managed & DCACHE_MOUNTED) {
826 struct vfsmount *mounted = lookup_mnt(path);
827 if (mounted) {
828 dput(path->dentry);
829 if (need_mntput)
830 mntput(path->mnt);
831 path->mnt = mounted;
832 path->dentry = dget(mounted->mnt_root);
833 need_mntput = true;
834 continue;
835 }
836
837 /* Something is mounted on this dentry in another
838 * namespace and/or whatever was mounted there in this
839 * namespace got unmounted before we managed to get the
840 * vfsmount_lock */
841 }
842
843 /* Handle an automount point */
844 if (managed & DCACHE_NEED_AUTOMOUNT) {
845 ret = follow_automount(path, flags, &need_mntput);
846 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100847 break;
David Howells9875cf82011-01-14 18:45:21 +0000848 continue;
849 }
850
851 /* We didn't change the current path point */
852 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 }
Al Viro8aef1882011-06-16 15:10:06 +0100854
855 if (need_mntput && path->mnt == mnt)
856 mntput(path->mnt);
857 if (ret == -EISDIR)
858 ret = 0;
859 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
861
David Howellscc53ce52011-01-14 18:45:26 +0000862int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
864 struct vfsmount *mounted;
865
Al Viro1c755af2009-04-18 14:06:57 -0400866 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400868 dput(path->dentry);
869 mntput(path->mnt);
870 path->mnt = mounted;
871 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return 1;
873 }
874 return 0;
875}
876
Ian Kent62a73752011-03-25 01:51:02 +0800877static inline bool managed_dentry_might_block(struct dentry *dentry)
878{
879 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
880 dentry->d_op->d_manage(dentry, true) < 0);
881}
882
David Howells9875cf82011-01-14 18:45:21 +0000883/*
Al Viro287548e2011-05-27 06:50:06 -0400884 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
885 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000886 */
887static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400888 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000889{
Ian Kent62a73752011-03-25 01:51:02 +0800890 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +0000891 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800892 /*
893 * Don't forget we might have a non-mountpoint managed dentry
894 * that wants to block transit.
895 */
Al Viro287548e2011-05-27 06:50:06 -0400896 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000897 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800898
899 if (!d_mountpoint(path->dentry))
900 break;
901
David Howells9875cf82011-01-14 18:45:21 +0000902 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
903 if (!mounted)
904 break;
905 path->mnt = mounted;
906 path->dentry = mounted->mnt_root;
907 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700908 /*
909 * Update the inode too. We don't need to re-check the
910 * dentry sequence number here after this d_inode read,
911 * because a mount-point is always pinned.
912 */
913 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000914 }
David Howells9875cf82011-01-14 18:45:21 +0000915 return true;
916}
917
Al Virodea39372011-05-27 06:53:39 -0400918static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400919{
Al Virodea39372011-05-27 06:53:39 -0400920 while (d_mountpoint(nd->path.dentry)) {
Al Viro287548e2011-05-27 06:50:06 -0400921 struct vfsmount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400922 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400923 if (!mounted)
924 break;
Al Virodea39372011-05-27 06:53:39 -0400925 nd->path.mnt = mounted;
926 nd->path.dentry = mounted->mnt_root;
927 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400928 }
929}
930
Nick Piggin31e6b012011-01-07 17:49:52 +1100931static int follow_dotdot_rcu(struct nameidata *nd)
932{
Nick Piggin31e6b012011-01-07 17:49:52 +1100933 set_root_rcu(nd);
934
David Howells9875cf82011-01-14 18:45:21 +0000935 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100936 if (nd->path.dentry == nd->root.dentry &&
937 nd->path.mnt == nd->root.mnt) {
938 break;
939 }
940 if (nd->path.dentry != nd->path.mnt->mnt_root) {
941 struct dentry *old = nd->path.dentry;
942 struct dentry *parent = old->d_parent;
943 unsigned seq;
944
945 seq = read_seqcount_begin(&parent->d_seq);
946 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500947 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100948 nd->path.dentry = parent;
949 nd->seq = seq;
950 break;
951 }
952 if (!follow_up_rcu(&nd->path))
953 break;
954 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100955 }
Al Virodea39372011-05-27 06:53:39 -0400956 follow_mount_rcu(nd);
957 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100958 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500959
960failed:
961 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500962 if (!(nd->flags & LOOKUP_ROOT))
963 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500964 rcu_read_unlock();
965 br_read_unlock(vfsmount_lock);
966 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100967}
968
David Howells9875cf82011-01-14 18:45:21 +0000969/*
David Howellscc53ce52011-01-14 18:45:26 +0000970 * Follow down to the covering mount currently visible to userspace. At each
971 * point, the filesystem owning that dentry may be queried as to whether the
972 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000973 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400974int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000975{
976 unsigned managed;
977 int ret;
978
979 while (managed = ACCESS_ONCE(path->dentry->d_flags),
980 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
981 /* Allow the filesystem to manage the transit without i_mutex
982 * being held.
983 *
984 * We indicate to the filesystem if someone is trying to mount
985 * something here. This gives autofs the chance to deny anyone
986 * other than its daemon the right to mount on its
987 * superstructure.
988 *
989 * The filesystem may sleep at this point.
990 */
991 if (managed & DCACHE_MANAGE_TRANSIT) {
992 BUG_ON(!path->dentry->d_op);
993 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000994 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400995 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000996 if (ret < 0)
997 return ret == -EISDIR ? 0 : ret;
998 }
999
1000 /* Transit to a mounted filesystem. */
1001 if (managed & DCACHE_MOUNTED) {
1002 struct vfsmount *mounted = lookup_mnt(path);
1003 if (!mounted)
1004 break;
1005 dput(path->dentry);
1006 mntput(path->mnt);
1007 path->mnt = mounted;
1008 path->dentry = dget(mounted->mnt_root);
1009 continue;
1010 }
1011
1012 /* Don't handle automount points here */
1013 break;
1014 }
1015 return 0;
1016}
1017
1018/*
David Howells9875cf82011-01-14 18:45:21 +00001019 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1020 */
1021static void follow_mount(struct path *path)
1022{
1023 while (d_mountpoint(path->dentry)) {
1024 struct vfsmount *mounted = lookup_mnt(path);
1025 if (!mounted)
1026 break;
1027 dput(path->dentry);
1028 mntput(path->mnt);
1029 path->mnt = mounted;
1030 path->dentry = dget(mounted->mnt_root);
1031 }
1032}
1033
Nick Piggin31e6b012011-01-07 17:49:52 +11001034static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
Al Viro2a737872009-04-07 11:49:53 -04001036 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001039 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Al Viro2a737872009-04-07 11:49:53 -04001041 if (nd->path.dentry == nd->root.dentry &&
1042 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 break;
1044 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001045 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001046 /* rare case of legitimate dget_parent()... */
1047 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 dput(old);
1049 break;
1050 }
Al Viro3088dd72010-01-30 15:47:29 -05001051 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
Al Viro79ed0222009-04-18 13:59:41 -04001054 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001055 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056}
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001059 * Allocate a dentry with name and parent, and perform a parent
1060 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1061 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1062 * have verified that no child exists while under i_mutex.
1063 */
1064static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1065 struct qstr *name, struct nameidata *nd)
1066{
1067 struct inode *inode = parent->d_inode;
1068 struct dentry *dentry;
1069 struct dentry *old;
1070
1071 /* Don't create child dentry for a dead directory. */
1072 if (unlikely(IS_DEADDIR(inode)))
1073 return ERR_PTR(-ENOENT);
1074
1075 dentry = d_alloc(parent, name);
1076 if (unlikely(!dentry))
1077 return ERR_PTR(-ENOMEM);
1078
1079 old = inode->i_op->lookup(inode, dentry, nd);
1080 if (unlikely(old)) {
1081 dput(dentry);
1082 dentry = old;
1083 }
1084 return dentry;
1085}
1086
1087/*
Josef Bacik44396f42011-05-31 11:58:49 -04001088 * We already have a dentry, but require a lookup to be performed on the parent
1089 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1090 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1091 * child exists while under i_mutex.
1092 */
1093static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1094 struct nameidata *nd)
1095{
1096 struct inode *inode = parent->d_inode;
1097 struct dentry *old;
1098
1099 /* Don't create child dentry for a dead directory. */
1100 if (unlikely(IS_DEADDIR(inode)))
1101 return ERR_PTR(-ENOENT);
1102
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1105 dput(dentry);
1106 dentry = old;
1107 }
1108 return dentry;
1109}
1110
1111/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 * It's more convoluted than I'd like it to be, but... it's still fairly
1113 * small and for now I'd prefer to have fast path as straight as possible.
1114 * It _is_ time-critical.
1115 */
1116static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001117 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118{
Jan Blunck4ac91372008-02-14 19:34:32 -08001119 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001120 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001121 int need_reval = 1;
1122 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001123 int err;
1124
Al Viro3cac2602009-08-13 18:27:43 +04001125 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001126 * Rename seqlock is not required here because in the off chance
1127 * of a false negative due to a concurrent rename, we're going to
1128 * do the non-racy lookup, below.
1129 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001130 if (nd->flags & LOOKUP_RCU) {
1131 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001132 *inode = nd->inode;
1133 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001134 if (!dentry)
1135 goto unlazy;
1136
Nick Piggin31e6b012011-01-07 17:49:52 +11001137 /* Memory barrier in read_seqcount_begin of child is enough */
1138 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1139 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001140 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001141
Al Viro24643082011-02-15 01:26:22 -05001142 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001143 status = d_revalidate(dentry, nd);
1144 if (unlikely(status <= 0)) {
1145 if (status != -ECHILD)
1146 need_reval = 0;
1147 goto unlazy;
1148 }
Al Viro24643082011-02-15 01:26:22 -05001149 }
Josef Bacik44396f42011-05-31 11:58:49 -04001150 if (unlikely(d_need_lookup(dentry)))
1151 goto unlazy;
Nick Piggin31e6b012011-01-07 17:49:52 +11001152 path->mnt = mnt;
1153 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001154 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1155 goto unlazy;
1156 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1157 goto unlazy;
1158 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001159unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001160 if (unlazy_walk(nd, dentry))
1161 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001162 } else {
1163 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001164 }
Al Viro5a18fff2011-03-11 04:44:53 -05001165
Josef Bacik44396f42011-05-31 11:58:49 -04001166 if (dentry && unlikely(d_need_lookup(dentry))) {
1167 dput(dentry);
1168 dentry = NULL;
1169 }
Al Viro5a18fff2011-03-11 04:44:53 -05001170retry:
1171 if (unlikely(!dentry)) {
1172 struct inode *dir = parent->d_inode;
1173 BUG_ON(nd->inode != dir);
1174
1175 mutex_lock(&dir->i_mutex);
1176 dentry = d_lookup(parent, name);
1177 if (likely(!dentry)) {
1178 dentry = d_alloc_and_lookup(parent, name, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1182 }
1183 /* known good */
1184 need_reval = 0;
1185 status = 1;
Josef Bacik44396f42011-05-31 11:58:49 -04001186 } else if (unlikely(d_need_lookup(dentry))) {
1187 dentry = d_inode_lookup(parent, dentry, nd);
1188 if (IS_ERR(dentry)) {
1189 mutex_unlock(&dir->i_mutex);
1190 return PTR_ERR(dentry);
1191 }
1192 /* known good */
1193 need_reval = 0;
1194 status = 1;
Al Viro5a18fff2011-03-11 04:44:53 -05001195 }
1196 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001197 }
Al Viro5a18fff2011-03-11 04:44:53 -05001198 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1199 status = d_revalidate(dentry, nd);
1200 if (unlikely(status <= 0)) {
1201 if (status < 0) {
1202 dput(dentry);
1203 return status;
1204 }
1205 if (!d_invalidate(dentry)) {
1206 dput(dentry);
1207 dentry = NULL;
1208 need_reval = 1;
1209 goto retry;
1210 }
1211 }
1212
David Howells9875cf82011-01-14 18:45:21 +00001213 path->mnt = mnt;
1214 path->dentry = dentry;
1215 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001216 if (unlikely(err < 0)) {
1217 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001218 return err;
Ian Kent89312212011-01-18 12:06:10 +08001219 }
David Howells9875cf82011-01-14 18:45:21 +00001220 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
Al Viro52094c82011-02-21 21:34:47 -05001224static inline int may_lookup(struct nameidata *nd)
1225{
1226 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001227 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001228 if (err != -ECHILD)
1229 return err;
Al Viro19660af2011-03-25 10:32:48 -04001230 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001231 return -ECHILD;
1232 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001233 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001234}
1235
Al Viro9856fa12011-03-04 14:22:06 -05001236static inline int handle_dots(struct nameidata *nd, int type)
1237{
1238 if (type == LAST_DOTDOT) {
1239 if (nd->flags & LOOKUP_RCU) {
1240 if (follow_dotdot_rcu(nd))
1241 return -ECHILD;
1242 } else
1243 follow_dotdot(nd);
1244 }
1245 return 0;
1246}
1247
Al Viro951361f2011-03-04 14:44:37 -05001248static void terminate_walk(struct nameidata *nd)
1249{
1250 if (!(nd->flags & LOOKUP_RCU)) {
1251 path_put(&nd->path);
1252 } else {
1253 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001254 if (!(nd->flags & LOOKUP_ROOT))
1255 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001256 rcu_read_unlock();
1257 br_read_unlock(vfsmount_lock);
1258 }
1259}
1260
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001261/*
1262 * Do we need to follow links? We _really_ want to be able
1263 * to do this check without having to look at inode->i_op,
1264 * so we keep a cache of "no, this doesn't need follow_link"
1265 * for the common case.
1266 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001267static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001268{
1269 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1270 if (likely(inode->i_op->follow_link))
1271 return follow;
1272
1273 /* This gets set once for the inode lifetime */
1274 spin_lock(&inode->i_lock);
1275 inode->i_opflags |= IOP_NOFOLLOW;
1276 spin_unlock(&inode->i_lock);
1277 }
1278 return 0;
1279}
1280
Al Viroce57dfc2011-03-13 19:58:58 -04001281static inline int walk_component(struct nameidata *nd, struct path *path,
1282 struct qstr *name, int type, int follow)
1283{
1284 struct inode *inode;
1285 int err;
1286 /*
1287 * "." and ".." are special - ".." especially so because it has
1288 * to be able to know about the current root directory and
1289 * parent relationships.
1290 */
1291 if (unlikely(type != LAST_NORM))
1292 return handle_dots(nd, type);
1293 err = do_lookup(nd, name, path, &inode);
1294 if (unlikely(err)) {
1295 terminate_walk(nd);
1296 return err;
1297 }
1298 if (!inode) {
1299 path_to_nameidata(path, nd);
1300 terminate_walk(nd);
1301 return -ENOENT;
1302 }
Linus Torvalds7813b942011-08-07 09:53:20 -07001303 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001304 if (nd->flags & LOOKUP_RCU) {
1305 if (unlikely(unlazy_walk(nd, path->dentry))) {
1306 terminate_walk(nd);
1307 return -ECHILD;
1308 }
1309 }
Al Viroce57dfc2011-03-13 19:58:58 -04001310 BUG_ON(inode != path->dentry->d_inode);
1311 return 1;
1312 }
1313 path_to_nameidata(path, nd);
1314 nd->inode = inode;
1315 return 0;
1316}
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318/*
Al Virob3563792011-03-14 21:54:55 -04001319 * This limits recursive symlink follows to 8, while
1320 * limiting consecutive symlinks to 40.
1321 *
1322 * Without that kind of total limit, nasty chains of consecutive
1323 * symlinks can cause almost arbitrarily long lookups.
1324 */
1325static inline int nested_symlink(struct path *path, struct nameidata *nd)
1326{
1327 int res;
1328
Al Virob3563792011-03-14 21:54:55 -04001329 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1330 path_put_conditional(path, nd);
1331 path_put(&nd->path);
1332 return -ELOOP;
1333 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001334 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001335
1336 nd->depth++;
1337 current->link_count++;
1338
1339 do {
1340 struct path link = *path;
1341 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001342
1343 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001344 if (!res)
1345 res = walk_component(nd, path, &nd->last,
1346 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001347 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001348 } while (res > 0);
1349
1350 current->link_count--;
1351 nd->depth--;
1352 return res;
1353}
1354
1355/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001356 * We really don't want to look at inode->i_op->lookup
1357 * when we don't have to. So we keep a cache bit in
1358 * the inode ->i_opflags field that says "yes, we can
1359 * do lookup on this inode".
1360 */
1361static inline int can_lookup(struct inode *inode)
1362{
1363 if (likely(inode->i_opflags & IOP_LOOKUP))
1364 return 1;
1365 if (likely(!inode->i_op->lookup))
1366 return 0;
1367
1368 /* We do this once for the lifetime of the inode */
1369 spin_lock(&inode->i_lock);
1370 inode->i_opflags |= IOP_LOOKUP;
1371 spin_unlock(&inode->i_lock);
1372 return 1;
1373}
1374
1375/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001377 * This is the basic name resolution function, turning a pathname into
1378 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001380 * Returns 0 and nd will have valid dentry and mnt on success.
1381 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 */
Al Viro6de88d72009-08-09 01:41:57 +04001383static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384{
1385 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388 while (*name=='/')
1389 name++;
1390 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001391 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 /* At this point we know we have a real path component. */
1394 for(;;) {
1395 unsigned long hash;
1396 struct qstr this;
1397 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001398 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Al Viro52094c82011-02-21 21:34:47 -05001400 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (err)
1402 break;
1403
1404 this.name = name;
1405 c = *(const unsigned char *)name;
1406
1407 hash = init_name_hash();
1408 do {
1409 name++;
1410 hash = partial_name_hash(c, hash);
1411 c = *(const unsigned char *)name;
1412 } while (c && (c != '/'));
1413 this.len = name - (const char *) this.name;
1414 this.hash = end_name_hash(hash);
1415
Al Virofe479a52011-02-22 15:10:03 -05001416 type = LAST_NORM;
1417 if (this.name[0] == '.') switch (this.len) {
1418 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001419 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001420 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001421 nd->flags |= LOOKUP_JUMPED;
1422 }
Al Virofe479a52011-02-22 15:10:03 -05001423 break;
1424 case 1:
1425 type = LAST_DOT;
1426 }
Al Viro5a202bc2011-03-08 14:17:44 -05001427 if (likely(type == LAST_NORM)) {
1428 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001429 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001430 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1431 err = parent->d_op->d_hash(parent, nd->inode,
1432 &this);
1433 if (err < 0)
1434 break;
1435 }
1436 }
Al Virofe479a52011-02-22 15:10:03 -05001437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 /* remove trailing slashes? */
1439 if (!c)
1440 goto last_component;
1441 while (*++name == '/');
1442 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001443 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
Al Viroce57dfc2011-03-13 19:58:58 -04001445 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1446 if (err < 0)
1447 return err;
Al Virofe479a52011-02-22 15:10:03 -05001448
Al Viroce57dfc2011-03-13 19:58:58 -04001449 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001450 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001452 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001453 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001454 if (can_lookup(nd->inode))
1455 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001457 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 /* here ends the main loop */
1459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460last_component:
Al Virob3563792011-03-14 21:54:55 -04001461 nd->last = this;
1462 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001463 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 }
Al Viro951361f2011-03-04 14:44:37 -05001465 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 return err;
1467}
1468
Al Viro70e9b352011-03-05 21:12:22 -05001469static int path_init(int dfd, const char *name, unsigned int flags,
1470 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001472 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001473 int fput_needed;
1474 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001477 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001479 if (flags & LOOKUP_ROOT) {
1480 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001481 if (*name) {
1482 if (!inode->i_op->lookup)
1483 return -ENOTDIR;
1484 retval = inode_permission(inode, MAY_EXEC);
1485 if (retval)
1486 return retval;
1487 }
Al Viro5b6ca022011-03-09 23:04:47 -05001488 nd->path = nd->root;
1489 nd->inode = inode;
1490 if (flags & LOOKUP_RCU) {
1491 br_read_lock(vfsmount_lock);
1492 rcu_read_lock();
1493 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1494 } else {
1495 path_get(&nd->path);
1496 }
1497 return 0;
1498 }
1499
Al Viro2a737872009-04-07 11:49:53 -04001500 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001503 if (flags & LOOKUP_RCU) {
1504 br_read_lock(vfsmount_lock);
1505 rcu_read_lock();
1506 set_root_rcu(nd);
1507 } else {
1508 set_root(nd);
1509 path_get(&nd->root);
1510 }
Al Viro2a737872009-04-07 11:49:53 -04001511 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001512 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001513 if (flags & LOOKUP_RCU) {
1514 struct fs_struct *fs = current->fs;
1515 unsigned seq;
1516
1517 br_read_lock(vfsmount_lock);
1518 rcu_read_lock();
1519
1520 do {
1521 seq = read_seqcount_begin(&fs->seq);
1522 nd->path = fs->pwd;
1523 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1524 } while (read_seqcount_retry(&fs->seq, seq));
1525 } else {
1526 get_fs_pwd(current->fs, &nd->path);
1527 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001528 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001529 struct dentry *dentry;
1530
Al Viro1abf0c72011-03-13 03:51:11 -04001531 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001532 retval = -EBADF;
1533 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001534 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001535
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001536 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001537
Al Virof52e0c12011-03-14 18:56:51 -04001538 if (*name) {
1539 retval = -ENOTDIR;
1540 if (!S_ISDIR(dentry->d_inode->i_mode))
1541 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001542
Al Viro4ad5abb2011-06-20 19:57:03 -04001543 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001544 if (retval)
1545 goto fput_fail;
1546 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001547
Jan Blunck5dd784d2008-02-14 19:34:38 -08001548 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001549 if (flags & LOOKUP_RCU) {
1550 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001551 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001552 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1553 br_read_lock(vfsmount_lock);
1554 rcu_read_lock();
1555 } else {
1556 path_get(&file->f_path);
1557 fput_light(file, fput_needed);
1558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 }
Al Viroe41f7d42011-02-22 14:02:58 -05001560
Nick Piggin31e6b012011-01-07 17:49:52 +11001561 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001562 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001563
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001564fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001565 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001566out_fail:
1567 return retval;
1568}
1569
Al Virobd92d7f2011-03-14 19:54:59 -04001570static inline int lookup_last(struct nameidata *nd, struct path *path)
1571{
1572 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1573 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1574
1575 nd->flags &= ~LOOKUP_PARENT;
1576 return walk_component(nd, path, &nd->last, nd->last_type,
1577 nd->flags & LOOKUP_FOLLOW);
1578}
1579
Al Viro9b4a9b12009-04-07 11:44:16 -04001580/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001581static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001582 unsigned int flags, struct nameidata *nd)
1583{
Al Viro70e9b352011-03-05 21:12:22 -05001584 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001585 struct path path;
1586 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001587
1588 /*
1589 * Path walking is largely split up into 2 different synchronisation
1590 * schemes, rcu-walk and ref-walk (explained in
1591 * Documentation/filesystems/path-lookup.txt). These share much of the
1592 * path walk code, but some things particularly setup, cleanup, and
1593 * following mounts are sufficiently divergent that functions are
1594 * duplicated. Typically there is a function foo(), and its RCU
1595 * analogue, foo_rcu().
1596 *
1597 * -ECHILD is the error number of choice (just to avoid clashes) that
1598 * is returned if some aspect of an rcu-walk fails. Such an error must
1599 * be handled by restarting a traditional ref-walk (which will always
1600 * be able to complete).
1601 */
Al Virobd92d7f2011-03-14 19:54:59 -04001602 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001603
Al Virobd92d7f2011-03-14 19:54:59 -04001604 if (unlikely(err))
1605 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001606
1607 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001608 err = link_path_walk(name, nd);
1609
1610 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001611 err = lookup_last(nd, &path);
1612 while (err > 0) {
1613 void *cookie;
1614 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001615 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001616 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001617 if (!err)
1618 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001619 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001620 }
1621 }
Al Viroee0827c2011-02-21 23:38:09 -05001622
Al Viro9f1fafe2011-03-25 11:00:12 -04001623 if (!err)
1624 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001625
1626 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1627 if (!nd->inode->i_op->lookup) {
1628 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001629 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001630 }
1631 }
Al Viro16c2cd72011-02-22 15:50:10 -05001632
Al Viro70e9b352011-03-05 21:12:22 -05001633 if (base)
1634 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001635
Al Viro5b6ca022011-03-09 23:04:47 -05001636 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001637 path_put(&nd->root);
1638 nd->root.mnt = NULL;
1639 }
Al Virobd92d7f2011-03-14 19:54:59 -04001640 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001641}
Nick Piggin31e6b012011-01-07 17:49:52 +11001642
Al Viroee0827c2011-02-21 23:38:09 -05001643static int do_path_lookup(int dfd, const char *name,
1644 unsigned int flags, struct nameidata *nd)
1645{
1646 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1647 if (unlikely(retval == -ECHILD))
1648 retval = path_lookupat(dfd, name, flags, nd);
1649 if (unlikely(retval == -ESTALE))
1650 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001651
1652 if (likely(!retval)) {
1653 if (unlikely(!audit_dummy_context())) {
1654 if (nd->path.dentry && nd->inode)
1655 audit_inode(name, nd->path.dentry);
1656 }
1657 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001658 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
Al Viroc9c6cac2011-02-16 15:15:47 -05001661int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001662{
Al Viroc9c6cac2011-02-16 15:15:47 -05001663 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001664}
1665
Al Virod1811462008-08-02 00:49:18 -04001666int kern_path(const char *name, unsigned int flags, struct path *path)
1667{
1668 struct nameidata nd;
1669 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1670 if (!res)
1671 *path = nd.path;
1672 return res;
1673}
1674
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001675/**
1676 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1677 * @dentry: pointer to dentry of the base directory
1678 * @mnt: pointer to vfs mount of the base directory
1679 * @name: pointer to file name
1680 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001681 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001682 */
1683int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1684 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001685 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001686{
Al Viroe0a01242011-06-27 17:00:37 -04001687 struct nameidata nd;
1688 int err;
1689 nd.root.dentry = dentry;
1690 nd.root.mnt = mnt;
1691 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001692 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001693 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1694 if (!err)
1695 *path = nd.path;
1696 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001697}
1698
Christoph Hellwigeead1912007-10-16 23:25:38 -07001699static struct dentry *__lookup_hash(struct qstr *name,
1700 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001702 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001703 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 int err;
1705
Al Viro4ad5abb2011-06-20 19:57:03 -04001706 err = inode_permission(inode, MAY_EXEC);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001707 if (err)
1708 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
1710 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001711 * Don't bother with __d_lookup: callers are for creat as
1712 * well as unlink, so a lot of the time it would cost
1713 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001714 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001715 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001716
Josef Bacik44396f42011-05-31 11:58:49 -04001717 if (dentry && d_need_lookup(dentry)) {
1718 /*
1719 * __lookup_hash is called with the parent dir's i_mutex already
1720 * held, so we are good to go here.
1721 */
1722 dentry = d_inode_lookup(base, dentry, nd);
1723 if (IS_ERR(dentry))
1724 return dentry;
1725 }
1726
Al Virod2d9e9f2011-06-20 10:55:26 -04001727 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1728 int status = d_revalidate(dentry, nd);
1729 if (unlikely(status <= 0)) {
1730 /*
1731 * The dentry failed validation.
1732 * If d_revalidate returned 0 attempt to invalidate
1733 * the dentry otherwise d_revalidate is asking us
1734 * to return a fail status.
1735 */
1736 if (status < 0) {
1737 dput(dentry);
1738 return ERR_PTR(status);
1739 } else if (!d_invalidate(dentry)) {
1740 dput(dentry);
1741 dentry = NULL;
1742 }
1743 }
1744 }
Al Viro6e6b1bd2009-08-13 23:38:37 +04001745
Nick Pigginbaa03892010-08-18 04:37:31 +10001746 if (!dentry)
1747 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 return dentry;
1750}
1751
James Morris057f6c02007-04-26 00:12:05 -07001752/*
1753 * Restricted form of lookup. Doesn't follow links, single-component only,
1754 * needs parent already locked. Doesn't follow mounts.
1755 * SMP-safe.
1756 */
Adrian Bunka244e162006-03-31 02:32:11 -08001757static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
Jan Blunck4ac91372008-02-14 19:34:32 -08001759 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760}
1761
Christoph Hellwigeead1912007-10-16 23:25:38 -07001762/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001763 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001764 * @name: pathname component to lookup
1765 * @base: base directory to lookup from
1766 * @len: maximum length @len should be interpreted to
1767 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001768 * Note that this routine is purely a helper for filesystem usage and should
1769 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001770 * nameidata argument is passed to the filesystem methods and a filesystem
1771 * using this helper needs to be prepared for that.
1772 */
James Morris057f6c02007-04-26 00:12:05 -07001773struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1774{
James Morris057f6c02007-04-26 00:12:05 -07001775 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001776 unsigned long hash;
1777 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001778
David Woodhouse2f9092e2009-04-20 23:18:37 +01001779 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1780
Al Viro6a96ba52011-03-07 23:49:20 -05001781 this.name = name;
1782 this.len = len;
1783 if (!len)
1784 return ERR_PTR(-EACCES);
1785
1786 hash = init_name_hash();
1787 while (len--) {
1788 c = *(const unsigned char *)name++;
1789 if (c == '/' || c == '\0')
1790 return ERR_PTR(-EACCES);
1791 hash = partial_name_hash(c, hash);
1792 }
1793 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001794 /*
1795 * See if the low-level filesystem might want
1796 * to use its own hash..
1797 */
1798 if (base->d_flags & DCACHE_OP_HASH) {
1799 int err = base->d_op->d_hash(base, base->d_inode, &this);
1800 if (err < 0)
1801 return ERR_PTR(err);
1802 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001803
Christoph Hellwig49705b72005-11-08 21:35:06 -08001804 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001805}
1806
Al Viro2d8f3032008-07-22 09:59:21 -04001807int user_path_at(int dfd, const char __user *name, unsigned flags,
1808 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809{
Al Viro2d8f3032008-07-22 09:59:21 -04001810 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001811 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001814
1815 BUG_ON(flags & LOOKUP_PARENT);
1816
1817 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001819 if (!err)
1820 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
1822 return err;
1823}
1824
Al Viro2ad94ae2008-07-21 09:32:51 -04001825static int user_path_parent(int dfd, const char __user *path,
1826 struct nameidata *nd, char **name)
1827{
1828 char *s = getname(path);
1829 int error;
1830
1831 if (IS_ERR(s))
1832 return PTR_ERR(s);
1833
1834 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1835 if (error)
1836 putname(s);
1837 else
1838 *name = s;
1839
1840 return error;
1841}
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843/*
1844 * It's inline, so penalty for filesystems that don't use sticky bit is
1845 * minimal.
1846 */
1847static inline int check_sticky(struct inode *dir, struct inode *inode)
1848{
David Howellsda9592e2008-11-14 10:39:05 +11001849 uid_t fsuid = current_fsuid();
1850
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 if (!(dir->i_mode & S_ISVTX))
1852 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001853 if (current_user_ns() != inode_userns(inode))
1854 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001855 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001857 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001859
1860other_userns:
1861 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862}
1863
1864/*
1865 * Check whether we can remove a link victim from directory dir, check
1866 * whether the type of victim is right.
1867 * 1. We can't do it if dir is read-only (done in permission())
1868 * 2. We should have write and exec permissions on dir
1869 * 3. We can't remove anything from append-only dir
1870 * 4. We can't do anything with immutable dir (done in permission())
1871 * 5. If the sticky bit on dir is set we should either
1872 * a. be owner of dir, or
1873 * b. be owner of victim, or
1874 * c. have CAP_FOWNER capability
1875 * 6. If the victim is append-only or immutable we can't do antyhing with
1876 * links pointing to it.
1877 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1878 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1879 * 9. We can't remove a root or mountpoint.
1880 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1881 * nfs_async_unlink().
1882 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001883static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884{
1885 int error;
1886
1887 if (!victim->d_inode)
1888 return -ENOENT;
1889
1890 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001891 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Al Virof419a2e2008-07-22 00:07:17 -04001893 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 if (error)
1895 return error;
1896 if (IS_APPEND(dir))
1897 return -EPERM;
1898 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001899 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 return -EPERM;
1901 if (isdir) {
1902 if (!S_ISDIR(victim->d_inode->i_mode))
1903 return -ENOTDIR;
1904 if (IS_ROOT(victim))
1905 return -EBUSY;
1906 } else if (S_ISDIR(victim->d_inode->i_mode))
1907 return -EISDIR;
1908 if (IS_DEADDIR(dir))
1909 return -ENOENT;
1910 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1911 return -EBUSY;
1912 return 0;
1913}
1914
1915/* Check whether we can create an object with dentry child in directory
1916 * dir.
1917 * 1. We can't do it if child already exists (open has special treatment for
1918 * this case, but since we are inlined it's OK)
1919 * 2. We can't do it if dir is read-only (done in permission())
1920 * 3. We should have write and exec permissions on dir
1921 * 4. We can't do it if dir is immutable (done in permission())
1922 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001923static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924{
1925 if (child->d_inode)
1926 return -EEXIST;
1927 if (IS_DEADDIR(dir))
1928 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001929 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930}
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932/*
1933 * p1 and p2 should be directories on the same fs.
1934 */
1935struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1936{
1937 struct dentry *p;
1938
1939 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001940 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 return NULL;
1942 }
1943
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001944 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001946 p = d_ancestor(p2, p1);
1947 if (p) {
1948 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1949 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1950 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 }
1952
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001953 p = d_ancestor(p1, p2);
1954 if (p) {
1955 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1956 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1957 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 }
1959
Ingo Molnarf2eace22006-07-03 00:25:05 -07001960 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1961 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 return NULL;
1963}
1964
1965void unlock_rename(struct dentry *p1, struct dentry *p2)
1966{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001967 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001969 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001970 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 }
1972}
1973
1974int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1975 struct nameidata *nd)
1976{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001977 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 if (error)
1980 return error;
1981
Al Viroacfa4382008-12-04 10:06:33 -05001982 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 return -EACCES; /* shouldn't it be ENOSYS? */
1984 mode &= S_IALLUGO;
1985 mode |= S_IFREG;
1986 error = security_inode_create(dir, dentry, mode);
1987 if (error)
1988 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001990 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001991 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return error;
1993}
1994
Al Viro73d049a2011-03-11 12:08:24 -05001995static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001997 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 struct inode *inode = dentry->d_inode;
1999 int error;
2000
Al Virobcda7652011-03-13 16:42:14 -04002001 /* O_PATH? */
2002 if (!acc_mode)
2003 return 0;
2004
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 if (!inode)
2006 return -ENOENT;
2007
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002008 switch (inode->i_mode & S_IFMT) {
2009 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002011 case S_IFDIR:
2012 if (acc_mode & MAY_WRITE)
2013 return -EISDIR;
2014 break;
2015 case S_IFBLK:
2016 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002017 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002019 /*FALLTHRU*/
2020 case S_IFIFO:
2021 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002023 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002024 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002025
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002026 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002027 if (error)
2028 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 /*
2031 * An append-only file must be opened in append mode for writing.
2032 */
2033 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002034 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002035 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002037 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 }
2039
2040 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002041 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002042 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 /*
2045 * Ensure there are no outstanding leases on the file.
2046 */
Al Virob65a9cf2009-12-16 06:27:40 -05002047 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002048}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Jeff Laytone1181ee2010-12-07 16:19:50 -05002050static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002051{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002052 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002053 struct inode *inode = path->dentry->d_inode;
2054 int error = get_write_access(inode);
2055 if (error)
2056 return error;
2057 /*
2058 * Refuse to truncate files with mandatory locks held on them.
2059 */
2060 error = locks_verify_locked(inode);
2061 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002062 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002063 if (!error) {
2064 error = do_truncate(path->dentry, 0,
2065 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002066 filp);
Al Viro7715b522009-12-16 03:54:00 -05002067 }
2068 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002069 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
2071
Dave Hansend57999e2008-02-15 14:37:27 -08002072static inline int open_to_namei_flags(int flag)
2073{
Al Viro8a5e9292011-06-25 19:15:54 -04002074 if ((flag & O_ACCMODE) == 3)
2075 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002076 return flag;
2077}
2078
Nick Piggin31e6b012011-01-07 17:49:52 +11002079/*
Al Virofe2d35f2011-03-05 22:58:25 -05002080 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002081 */
Al Virofb1cc552009-12-24 01:58:28 -05002082static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002083 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002084{
Al Viroa1e28032009-12-24 02:12:06 -05002085 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002086 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002087 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002088 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002089 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002090 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002091 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002092 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002093
Al Viroc3e380b2011-02-23 13:39:45 -05002094 nd->flags &= ~LOOKUP_PARENT;
2095 nd->flags |= op->intent;
2096
Al Viro1f36f772009-12-26 10:56:19 -05002097 switch (nd->last_type) {
2098 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002099 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002100 error = handle_dots(nd, nd->last_type);
2101 if (error)
2102 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002103 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002104 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002105 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002106 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002107 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002108 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002109 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002110 error = -EISDIR;
2111 goto exit;
2112 }
2113 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002114 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002115 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002116 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002117 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002118 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002119 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002120 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002121
Al Viroca344a892011-03-09 00:36:45 -05002122 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002123 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002124 if (nd->last.name[nd->last.len])
2125 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002126 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2127 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002128 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002129 error = walk_component(nd, path, &nd->last, LAST_NORM,
2130 !symlink_ok);
2131 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002132 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002133 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002134 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002135 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002136 error = complete_walk(nd);
2137 if (error)
2138 return ERR_PTR(-ECHILD);
Al Virofe2d35f2011-03-05 22:58:25 -05002139
2140 error = -ENOTDIR;
2141 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002142 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002143 goto exit;
2144 }
2145 audit_inode(pathname, nd->path.dentry);
2146 goto ok;
2147 }
2148
2149 /* create side of things */
Al Viro9f1fafe2011-03-25 11:00:12 -04002150 error = complete_walk(nd);
2151 if (error)
2152 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002153
2154 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002155 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002156 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002157 if (nd->last.name[nd->last.len])
2158 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002159
Al Viroa1e28032009-12-24 02:12:06 -05002160 mutex_lock(&dir->d_inode->i_mutex);
2161
Al Viro6c0d46c2011-03-09 00:59:59 -05002162 dentry = lookup_hash(nd);
2163 error = PTR_ERR(dentry);
2164 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002165 mutex_unlock(&dir->d_inode->i_mutex);
2166 goto exit;
2167 }
2168
Al Viro6c0d46c2011-03-09 00:59:59 -05002169 path->dentry = dentry;
2170 path->mnt = nd->path.mnt;
2171
Al Virofb1cc552009-12-24 01:58:28 -05002172 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002173 if (!dentry->d_inode) {
2174 int mode = op->mode;
2175 if (!IS_POSIXACL(dir->d_inode))
2176 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002177 /*
2178 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002179 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002180 * the time when the file is created and when
2181 * a permanent write count is taken through
2182 * the 'struct file' in nameidata_to_filp().
2183 */
2184 error = mnt_want_write(nd->path.mnt);
2185 if (error)
2186 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002187 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002188 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002189 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002190 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002191 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002192 error = security_path_mknod(&nd->path, dentry, mode, 0);
2193 if (error)
2194 goto exit_mutex_unlock;
2195 error = vfs_create(dir->d_inode, dentry, mode, nd);
2196 if (error)
2197 goto exit_mutex_unlock;
2198 mutex_unlock(&dir->d_inode->i_mutex);
2199 dput(nd->path.dentry);
2200 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002201 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002202 }
2203
2204 /*
2205 * It already exists.
2206 */
2207 mutex_unlock(&dir->d_inode->i_mutex);
2208 audit_inode(pathname, path->dentry);
2209
2210 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002211 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002212 goto exit_dput;
2213
David Howells9875cf82011-01-14 18:45:21 +00002214 error = follow_managed(path, nd->flags);
2215 if (error < 0)
2216 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002217
2218 error = -ENOENT;
2219 if (!path->dentry->d_inode)
2220 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002221
2222 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002223 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002224
2225 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002226 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002227 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002228 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002229 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002230ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002231 if (!S_ISREG(nd->inode->i_mode))
2232 will_truncate = 0;
2233
Al Viro0f9d1a12011-03-09 00:13:14 -05002234 if (will_truncate) {
2235 error = mnt_want_write(nd->path.mnt);
2236 if (error)
2237 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002238 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002239 }
Al Viroca344a892011-03-09 00:36:45 -05002240common:
Al Virobcda7652011-03-13 16:42:14 -04002241 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002242 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002243 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002244 filp = nameidata_to_filp(nd);
2245 if (!IS_ERR(filp)) {
2246 error = ima_file_check(filp, op->acc_mode);
2247 if (error) {
2248 fput(filp);
2249 filp = ERR_PTR(error);
2250 }
2251 }
2252 if (!IS_ERR(filp)) {
2253 if (will_truncate) {
2254 error = handle_truncate(filp);
2255 if (error) {
2256 fput(filp);
2257 filp = ERR_PTR(error);
2258 }
2259 }
2260 }
Al Viroca344a892011-03-09 00:36:45 -05002261out:
2262 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002263 mnt_drop_write(nd->path.mnt);
2264 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002265 return filp;
2266
2267exit_mutex_unlock:
2268 mutex_unlock(&dir->d_inode->i_mutex);
2269exit_dput:
2270 path_put_conditional(path, nd);
2271exit:
Al Viroca344a892011-03-09 00:36:45 -05002272 filp = ERR_PTR(error);
2273 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002274}
2275
Al Viro13aab422011-02-23 17:54:08 -05002276static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002277 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278{
Al Virofe2d35f2011-03-05 22:58:25 -05002279 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002280 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002281 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002282 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002283
2284 filp = get_empty_filp();
2285 if (!filp)
2286 return ERR_PTR(-ENFILE);
2287
Al Viro47c805d2011-02-23 17:44:09 -05002288 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002289 nd->intent.open.file = filp;
2290 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2291 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002292
Al Viro73d049a2011-03-11 12:08:24 -05002293 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002294 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002295 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002296
Al Virofe2d35f2011-03-05 22:58:25 -05002297 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002298 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002299 if (unlikely(error))
2300 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Al Viro73d049a2011-03-11 12:08:24 -05002302 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002303 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002304 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002305 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002306 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002307 path_put_conditional(&path, nd);
2308 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002309 filp = ERR_PTR(-ELOOP);
2310 break;
2311 }
Al Viro73d049a2011-03-11 12:08:24 -05002312 nd->flags |= LOOKUP_PARENT;
2313 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002314 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002315 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002316 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002317 else
Al Viro73d049a2011-03-11 12:08:24 -05002318 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002319 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002320 }
Al Viro10fa8e62009-12-26 07:09:49 -05002321out:
Al Viro73d049a2011-03-11 12:08:24 -05002322 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2323 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002324 if (base)
2325 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002326 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002327 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
Nick Piggin31e6b012011-01-07 17:49:52 +11002329out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002330 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002331 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332}
2333
Al Viro13aab422011-02-23 17:54:08 -05002334struct file *do_filp_open(int dfd, const char *pathname,
2335 const struct open_flags *op, int flags)
2336{
Al Viro73d049a2011-03-11 12:08:24 -05002337 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002338 struct file *filp;
2339
Al Viro73d049a2011-03-11 12:08:24 -05002340 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002341 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002342 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002343 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002344 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002345 return filp;
2346}
2347
Al Viro73d049a2011-03-11 12:08:24 -05002348struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2349 const char *name, const struct open_flags *op, int flags)
2350{
2351 struct nameidata nd;
2352 struct file *file;
2353
2354 nd.root.mnt = mnt;
2355 nd.root.dentry = dentry;
2356
2357 flags |= LOOKUP_ROOT;
2358
Al Virobcda7652011-03-13 16:42:14 -04002359 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002360 return ERR_PTR(-ELOOP);
2361
2362 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2363 if (unlikely(file == ERR_PTR(-ECHILD)))
2364 file = path_openat(-1, name, &nd, op, flags);
2365 if (unlikely(file == ERR_PTR(-ESTALE)))
2366 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2367 return file;
2368}
2369
Al Viroed75e952011-06-27 16:53:43 -04002370struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002372 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002373 struct nameidata nd;
2374 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2375 if (error)
2376 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002378 /*
2379 * Yucky last component or no last component at all?
2380 * (foo/., foo/.., /////)
2381 */
Al Viroed75e952011-06-27 16:53:43 -04002382 if (nd.last_type != LAST_NORM)
2383 goto out;
2384 nd.flags &= ~LOOKUP_PARENT;
2385 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2386 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002387
2388 /*
2389 * Do the final lookup.
2390 */
Al Viroed75e952011-06-27 16:53:43 -04002391 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2392 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 if (IS_ERR(dentry))
2394 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002395
Al Viroe9baf6e2008-05-15 04:49:12 -04002396 if (dentry->d_inode)
2397 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002398 /*
2399 * Special case - lookup gave negative, but... we had foo/bar/
2400 * From the vfs_mknod() POV we just have a negative dentry -
2401 * all is fine. Let's be bastards - you had / on the end, you've
2402 * been asking for (non-existent) directory. -ENOENT for you.
2403 */
Al Viroed75e952011-06-27 16:53:43 -04002404 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002405 dput(dentry);
2406 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002407 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002408 }
Al Viroed75e952011-06-27 16:53:43 -04002409 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002411eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002413 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414fail:
Al Viroed75e952011-06-27 16:53:43 -04002415 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2416out:
2417 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 return dentry;
2419}
Al Virodae6ad82011-06-26 11:50:15 -04002420EXPORT_SYMBOL(kern_path_create);
2421
2422struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2423{
2424 char *tmp = getname(pathname);
2425 struct dentry *res;
2426 if (IS_ERR(tmp))
2427 return ERR_CAST(tmp);
2428 res = kern_path_create(dfd, tmp, path, is_dir);
2429 putname(tmp);
2430 return res;
2431}
2432EXPORT_SYMBOL(user_path_create);
2433
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2435{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002436 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
2438 if (error)
2439 return error;
2440
Serge E. Hallyne795b712011-03-23 16:43:25 -07002441 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2442 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 return -EPERM;
2444
Al Viroacfa4382008-12-04 10:06:33 -05002445 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 return -EPERM;
2447
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002448 error = devcgroup_inode_mknod(mode, dev);
2449 if (error)
2450 return error;
2451
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 error = security_inode_mknod(dir, dentry, mode, dev);
2453 if (error)
2454 return error;
2455
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002457 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002458 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 return error;
2460}
2461
Dave Hansen463c3192008-02-15 14:37:57 -08002462static int may_mknod(mode_t mode)
2463{
2464 switch (mode & S_IFMT) {
2465 case S_IFREG:
2466 case S_IFCHR:
2467 case S_IFBLK:
2468 case S_IFIFO:
2469 case S_IFSOCK:
2470 case 0: /* zero mode translates to S_IFREG */
2471 return 0;
2472 case S_IFDIR:
2473 return -EPERM;
2474 default:
2475 return -EINVAL;
2476 }
2477}
2478
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002479SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2480 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481{
Al Viro2ad94ae2008-07-21 09:32:51 -04002482 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002483 struct path path;
2484 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485
2486 if (S_ISDIR(mode))
2487 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Al Virodae6ad82011-06-26 11:50:15 -04002489 dentry = user_path_create(dfd, filename, &path, 0);
2490 if (IS_ERR(dentry))
2491 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002492
Al Virodae6ad82011-06-26 11:50:15 -04002493 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002494 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002495 error = may_mknod(mode);
2496 if (error)
2497 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002498 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002499 if (error)
2500 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002501 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002502 if (error)
2503 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002504 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002506 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 break;
2508 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002509 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 new_decode_dev(dev));
2511 break;
2512 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002513 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002516out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002517 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002518out_dput:
2519 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002520 mutex_unlock(&path.dentry->d_inode->i_mutex);
2521 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
2523 return error;
2524}
2525
Heiko Carstens3480b252009-01-14 14:14:16 +01002526SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002527{
2528 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2529}
2530
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2532{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002533 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 if (error)
2536 return error;
2537
Al Viroacfa4382008-12-04 10:06:33 -05002538 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 return -EPERM;
2540
2541 mode &= (S_IRWXUGO|S_ISVTX);
2542 error = security_inode_mkdir(dir, dentry, mode);
2543 if (error)
2544 return error;
2545
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002547 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002548 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 return error;
2550}
2551
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002552SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553{
Dave Hansen6902d922006-09-30 23:29:01 -07002554 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002555 struct path path;
2556 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
Al Virodae6ad82011-06-26 11:50:15 -04002558 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002559 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002560 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002561
Al Virodae6ad82011-06-26 11:50:15 -04002562 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002563 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002564 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002565 if (error)
2566 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002567 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002568 if (error)
2569 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002570 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002571out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002572 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002573out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002574 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002575 mutex_unlock(&path.dentry->d_inode->i_mutex);
2576 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 return error;
2578}
2579
Heiko Carstens3cdad422009-01-14 14:14:22 +01002580SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002581{
2582 return sys_mkdirat(AT_FDCWD, pathname, mode);
2583}
2584
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585/*
Sage Weila71905f2011-05-24 13:06:08 -07002586 * The dentry_unhash() helper will try to drop the dentry early: we
2587 * should have a usage count of 2 if we're the only user of this
2588 * dentry, and if that is true (possibly after pruning the dcache),
2589 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 *
2591 * A low-level filesystem can, if it choses, legally
2592 * do a
2593 *
2594 * if (!d_unhashed(dentry))
2595 * return -EBUSY;
2596 *
2597 * if it cannot handle the case of removing a directory
2598 * that is still in use by something else..
2599 */
2600void dentry_unhash(struct dentry *dentry)
2601{
Vasily Averindc168422006-12-06 20:37:07 -08002602 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002604 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 __d_drop(dentry);
2606 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607}
2608
2609int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2610{
2611 int error = may_delete(dir, dentry, 1);
2612
2613 if (error)
2614 return error;
2615
Al Viroacfa4382008-12-04 10:06:33 -05002616 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 return -EPERM;
2618
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002619 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Sage Weil912dbc12011-05-24 13:06:11 -07002621 error = -EBUSY;
2622 if (d_mountpoint(dentry))
2623 goto out;
2624
2625 error = security_inode_rmdir(dir, dentry);
2626 if (error)
2627 goto out;
2628
Sage Weil3cebde22011-05-29 21:20:59 -07002629 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002630 error = dir->i_op->rmdir(dir, dentry);
2631 if (error)
2632 goto out;
2633
2634 dentry->d_inode->i_flags |= S_DEAD;
2635 dont_mount(dentry);
2636
2637out:
2638 mutex_unlock(&dentry->d_inode->i_mutex);
2639 if (!error)
2640 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 return error;
2642}
2643
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002644static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645{
2646 int error = 0;
2647 char * name;
2648 struct dentry *dentry;
2649 struct nameidata nd;
2650
Al Viro2ad94ae2008-07-21 09:32:51 -04002651 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002653 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
2655 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002656 case LAST_DOTDOT:
2657 error = -ENOTEMPTY;
2658 goto exit1;
2659 case LAST_DOT:
2660 error = -EINVAL;
2661 goto exit1;
2662 case LAST_ROOT:
2663 error = -EBUSY;
2664 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002666
2667 nd.flags &= ~LOOKUP_PARENT;
2668
Jan Blunck4ac91372008-02-14 19:34:32 -08002669 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002670 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002672 if (IS_ERR(dentry))
2673 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002674 if (!dentry->d_inode) {
2675 error = -ENOENT;
2676 goto exit3;
2677 }
Dave Hansen06227532008-02-15 14:37:34 -08002678 error = mnt_want_write(nd.path.mnt);
2679 if (error)
2680 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002681 error = security_path_rmdir(&nd.path, dentry);
2682 if (error)
2683 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002684 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002685exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002686 mnt_drop_write(nd.path.mnt);
2687exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002688 dput(dentry);
2689exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002690 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002692 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 putname(name);
2694 return error;
2695}
2696
Heiko Carstens3cdad422009-01-14 14:14:22 +01002697SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002698{
2699 return do_rmdir(AT_FDCWD, pathname);
2700}
2701
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702int vfs_unlink(struct inode *dir, struct dentry *dentry)
2703{
2704 int error = may_delete(dir, dentry, 0);
2705
2706 if (error)
2707 return error;
2708
Al Viroacfa4382008-12-04 10:06:33 -05002709 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 return -EPERM;
2711
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002712 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 if (d_mountpoint(dentry))
2714 error = -EBUSY;
2715 else {
2716 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002717 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002719 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002720 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002723 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724
2725 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2726 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002727 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 }
Robert Love0eeca282005-07-12 17:06:03 -04002730
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 return error;
2732}
2733
2734/*
2735 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002736 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 * writeout happening, and we don't want to prevent access to the directory
2738 * while waiting on the I/O.
2739 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002740static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741{
Al Viro2ad94ae2008-07-21 09:32:51 -04002742 int error;
2743 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 struct dentry *dentry;
2745 struct nameidata nd;
2746 struct inode *inode = NULL;
2747
Al Viro2ad94ae2008-07-21 09:32:51 -04002748 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002750 return error;
2751
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 error = -EISDIR;
2753 if (nd.last_type != LAST_NORM)
2754 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002755
2756 nd.flags &= ~LOOKUP_PARENT;
2757
Jan Blunck4ac91372008-02-14 19:34:32 -08002758 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002759 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 error = PTR_ERR(dentry);
2761 if (!IS_ERR(dentry)) {
2762 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002763 if (nd.last.name[nd.last.len])
2764 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002766 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002767 goto slashes;
2768 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002769 error = mnt_want_write(nd.path.mnt);
2770 if (error)
2771 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002772 error = security_path_unlink(&nd.path, dentry);
2773 if (error)
2774 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002775 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002776exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002777 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 exit2:
2779 dput(dentry);
2780 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002781 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 if (inode)
2783 iput(inode); /* truncate the inode here */
2784exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002785 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 putname(name);
2787 return error;
2788
2789slashes:
2790 error = !dentry->d_inode ? -ENOENT :
2791 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2792 goto exit2;
2793}
2794
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002795SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002796{
2797 if ((flag & ~AT_REMOVEDIR) != 0)
2798 return -EINVAL;
2799
2800 if (flag & AT_REMOVEDIR)
2801 return do_rmdir(dfd, pathname);
2802
2803 return do_unlinkat(dfd, pathname);
2804}
2805
Heiko Carstens3480b252009-01-14 14:14:16 +01002806SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002807{
2808 return do_unlinkat(AT_FDCWD, pathname);
2809}
2810
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002811int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002813 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
2815 if (error)
2816 return error;
2817
Al Viroacfa4382008-12-04 10:06:33 -05002818 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 return -EPERM;
2820
2821 error = security_inode_symlink(dir, dentry, oldname);
2822 if (error)
2823 return error;
2824
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002826 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002827 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 return error;
2829}
2830
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002831SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2832 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833{
Al Viro2ad94ae2008-07-21 09:32:51 -04002834 int error;
2835 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07002836 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002837 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838
2839 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002840 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002842
Al Virodae6ad82011-06-26 11:50:15 -04002843 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07002844 error = PTR_ERR(dentry);
2845 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002846 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07002847
Al Virodae6ad82011-06-26 11:50:15 -04002848 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002849 if (error)
2850 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002851 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002852 if (error)
2853 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002854 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002855out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002856 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002857out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002858 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002859 mutex_unlock(&path.dentry->d_inode->i_mutex);
2860 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07002861out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 putname(from);
2863 return error;
2864}
2865
Heiko Carstens3480b252009-01-14 14:14:16 +01002866SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002867{
2868 return sys_symlinkat(oldname, AT_FDCWD, newname);
2869}
2870
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2872{
2873 struct inode *inode = old_dentry->d_inode;
2874 int error;
2875
2876 if (!inode)
2877 return -ENOENT;
2878
Miklos Szeredia95164d2008-07-30 15:08:48 +02002879 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 if (error)
2881 return error;
2882
2883 if (dir->i_sb != inode->i_sb)
2884 return -EXDEV;
2885
2886 /*
2887 * A link to an append-only or immutable file cannot be created.
2888 */
2889 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2890 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002891 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002893 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 return -EPERM;
2895
2896 error = security_inode_link(old_dentry, dir, new_dentry);
2897 if (error)
2898 return error;
2899
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002900 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302901 /* Make sure we don't allow creating hardlink to an unlinked file */
2902 if (inode->i_nlink == 0)
2903 error = -ENOENT;
2904 else
2905 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002906 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002907 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002908 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 return error;
2910}
2911
2912/*
2913 * Hardlinks are often used in delicate situations. We avoid
2914 * security-related surprises by not following symlinks on the
2915 * newname. --KAB
2916 *
2917 * We don't follow them on the oldname either to be compatible
2918 * with linux 2.0, and to avoid hard-linking to directories
2919 * and other special files. --ADM
2920 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002921SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2922 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923{
2924 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002925 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302926 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302929 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002930 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302931 /*
2932 * To use null names we require CAP_DAC_READ_SEARCH
2933 * This ensures that not everyone will be able to create
2934 * handlink using the passed filedescriptor.
2935 */
2936 if (flags & AT_EMPTY_PATH) {
2937 if (!capable(CAP_DAC_READ_SEARCH))
2938 return -ENOENT;
2939 how = LOOKUP_EMPTY;
2940 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002941
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302942 if (flags & AT_SYMLINK_FOLLOW)
2943 how |= LOOKUP_FOLLOW;
2944
2945 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002947 return error;
2948
Al Virodae6ad82011-06-26 11:50:15 -04002949 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002951 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002952 goto out;
2953
2954 error = -EXDEV;
2955 if (old_path.mnt != new_path.mnt)
2956 goto out_dput;
2957 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002958 if (error)
2959 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002960 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002961 if (error)
2962 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002963 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002964out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002965 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002966out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002967 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002968 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2969 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970out:
Al Viro2d8f3032008-07-22 09:59:21 -04002971 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972
2973 return error;
2974}
2975
Heiko Carstens3480b252009-01-14 14:14:16 +01002976SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002977{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002978 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002979}
2980
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981/*
2982 * The worst of all namespace operations - renaming directory. "Perverted"
2983 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2984 * Problems:
2985 * a) we can get into loop creation. Check is done in is_subdir().
2986 * b) race potential - two innocent renames can create a loop together.
2987 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002988 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 * story.
2990 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002991 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 * whether the target exists). Solution: try to be smart with locking
2993 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002994 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 * move will be locked. Thus we can rank directories by the tree
2996 * (ancestors first) and rank all non-directories after them.
2997 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002998 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 * HOWEVER, it relies on the assumption that any object with ->lookup()
3000 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3001 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003002 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003003 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003005 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 * locking].
3007 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003008static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3009 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010{
3011 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003012 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013
3014 /*
3015 * If we are going to change the parent - check write permissions,
3016 * we'll need to flip '..'.
3017 */
3018 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003019 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 if (error)
3021 return error;
3022 }
3023
3024 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3025 if (error)
3026 return error;
3027
Al Virod83c49f2010-04-30 17:17:09 -04003028 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003029 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003030
3031 error = -EBUSY;
3032 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3033 goto out;
3034
Sage Weil3cebde22011-05-29 21:20:59 -07003035 if (target)
3036 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003037 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3038 if (error)
3039 goto out;
3040
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003042 target->i_flags |= S_DEAD;
3043 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 }
Sage Weil9055cba2011-05-24 13:06:12 -07003045out:
3046 if (target)
3047 mutex_unlock(&target->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003048 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003049 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3050 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 return error;
3052}
3053
Adrian Bunk75c96f82005-05-05 16:16:09 -07003054static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3055 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056{
Sage Weil51892bb2011-05-24 13:06:13 -07003057 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 int error;
3059
3060 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3061 if (error)
3062 return error;
3063
3064 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003066 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003067
3068 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003070 goto out;
3071
3072 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3073 if (error)
3074 goto out;
3075
3076 if (target)
3077 dont_mount(new_dentry);
3078 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3079 d_move(old_dentry, new_dentry);
3080out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003082 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 dput(new_dentry);
3084 return error;
3085}
3086
3087int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3088 struct inode *new_dir, struct dentry *new_dentry)
3089{
3090 int error;
3091 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003092 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
3094 if (old_dentry->d_inode == new_dentry->d_inode)
3095 return 0;
3096
3097 error = may_delete(old_dir, old_dentry, is_dir);
3098 if (error)
3099 return error;
3100
3101 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003102 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 else
3104 error = may_delete(new_dir, new_dentry, is_dir);
3105 if (error)
3106 return error;
3107
Al Viroacfa4382008-12-04 10:06:33 -05003108 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 return -EPERM;
3110
Robert Love0eeca282005-07-12 17:06:03 -04003111 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3112
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 if (is_dir)
3114 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3115 else
3116 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003117 if (!error)
3118 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003119 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003120 fsnotify_oldname_free(old_name);
3121
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 return error;
3123}
3124
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003125SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3126 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127{
Al Viro2ad94ae2008-07-21 09:32:51 -04003128 struct dentry *old_dir, *new_dir;
3129 struct dentry *old_dentry, *new_dentry;
3130 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003132 char *from;
3133 char *to;
3134 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135
Al Viro2ad94ae2008-07-21 09:32:51 -04003136 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 if (error)
3138 goto exit;
3139
Al Viro2ad94ae2008-07-21 09:32:51 -04003140 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 if (error)
3142 goto exit1;
3143
3144 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003145 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 goto exit2;
3147
Jan Blunck4ac91372008-02-14 19:34:32 -08003148 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 error = -EBUSY;
3150 if (oldnd.last_type != LAST_NORM)
3151 goto exit2;
3152
Jan Blunck4ac91372008-02-14 19:34:32 -08003153 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 if (newnd.last_type != LAST_NORM)
3155 goto exit2;
3156
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003157 oldnd.flags &= ~LOOKUP_PARENT;
3158 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003159 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003160
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 trap = lock_rename(new_dir, old_dir);
3162
Christoph Hellwig49705b72005-11-08 21:35:06 -08003163 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 error = PTR_ERR(old_dentry);
3165 if (IS_ERR(old_dentry))
3166 goto exit3;
3167 /* source must exist */
3168 error = -ENOENT;
3169 if (!old_dentry->d_inode)
3170 goto exit4;
3171 /* unless the source is a directory trailing slashes give -ENOTDIR */
3172 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3173 error = -ENOTDIR;
3174 if (oldnd.last.name[oldnd.last.len])
3175 goto exit4;
3176 if (newnd.last.name[newnd.last.len])
3177 goto exit4;
3178 }
3179 /* source should not be ancestor of target */
3180 error = -EINVAL;
3181 if (old_dentry == trap)
3182 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003183 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 error = PTR_ERR(new_dentry);
3185 if (IS_ERR(new_dentry))
3186 goto exit4;
3187 /* target should not be an ancestor of source */
3188 error = -ENOTEMPTY;
3189 if (new_dentry == trap)
3190 goto exit5;
3191
Dave Hansen9079b1e2008-02-15 14:37:49 -08003192 error = mnt_want_write(oldnd.path.mnt);
3193 if (error)
3194 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003195 error = security_path_rename(&oldnd.path, old_dentry,
3196 &newnd.path, new_dentry);
3197 if (error)
3198 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 error = vfs_rename(old_dir->d_inode, old_dentry,
3200 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003201exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003202 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203exit5:
3204 dput(new_dentry);
3205exit4:
3206 dput(old_dentry);
3207exit3:
3208 unlock_rename(new_dir, old_dir);
3209exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003210 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003211 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003213 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003215exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 return error;
3217}
3218
Heiko Carstensa26eab22009-01-14 14:14:17 +01003219SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003220{
3221 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3222}
3223
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3225{
3226 int len;
3227
3228 len = PTR_ERR(link);
3229 if (IS_ERR(link))
3230 goto out;
3231
3232 len = strlen(link);
3233 if (len > (unsigned) buflen)
3234 len = buflen;
3235 if (copy_to_user(buffer, link, len))
3236 len = -EFAULT;
3237out:
3238 return len;
3239}
3240
3241/*
3242 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3243 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3244 * using) it for any given inode is up to filesystem.
3245 */
3246int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3247{
3248 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003249 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003250 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003251
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003253 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003254 if (IS_ERR(cookie))
3255 return PTR_ERR(cookie);
3256
3257 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3258 if (dentry->d_inode->i_op->put_link)
3259 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3260 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261}
3262
3263int vfs_follow_link(struct nameidata *nd, const char *link)
3264{
3265 return __vfs_follow_link(nd, link);
3266}
3267
3268/* get the link contents into pagecache */
3269static char *page_getlink(struct dentry * dentry, struct page **ppage)
3270{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003271 char *kaddr;
3272 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003274 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003276 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003278 kaddr = kmap(page);
3279 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3280 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281}
3282
3283int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3284{
3285 struct page *page = NULL;
3286 char *s = page_getlink(dentry, &page);
3287 int res = vfs_readlink(dentry,buffer,buflen,s);
3288 if (page) {
3289 kunmap(page);
3290 page_cache_release(page);
3291 }
3292 return res;
3293}
3294
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003295void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003297 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003299 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300}
3301
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003302void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003304 struct page *page = cookie;
3305
3306 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 kunmap(page);
3308 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 }
3310}
3311
Nick Piggin54566b22009-01-04 12:00:53 -08003312/*
3313 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3314 */
3315int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316{
3317 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003318 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003319 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003320 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003322 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3323 if (nofs)
3324 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325
NeilBrown7e53cac2006-03-25 03:07:57 -08003326retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003327 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003328 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003330 goto fail;
3331
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 kaddr = kmap_atomic(page, KM_USER0);
3333 memcpy(kaddr, symname, len-1);
3334 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003335
3336 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3337 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 if (err < 0)
3339 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003340 if (err < len-1)
3341 goto retry;
3342
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 mark_inode_dirty(inode);
3344 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345fail:
3346 return err;
3347}
3348
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003349int page_symlink(struct inode *inode, const char *symname, int len)
3350{
3351 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003352 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003353}
3354
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003355const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 .readlink = generic_readlink,
3357 .follow_link = page_follow_link_light,
3358 .put_link = page_put_link,
3359};
3360
Al Viro2d8f3032008-07-22 09:59:21 -04003361EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003362EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363EXPORT_SYMBOL(follow_down);
3364EXPORT_SYMBOL(follow_up);
3365EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3366EXPORT_SYMBOL(getname);
3367EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368EXPORT_SYMBOL(lookup_one_len);
3369EXPORT_SYMBOL(page_follow_link_light);
3370EXPORT_SYMBOL(page_put_link);
3371EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003372EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373EXPORT_SYMBOL(page_symlink);
3374EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003375EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003376EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003377EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378EXPORT_SYMBOL(unlock_rename);
3379EXPORT_SYMBOL(vfs_create);
3380EXPORT_SYMBOL(vfs_follow_link);
3381EXPORT_SYMBOL(vfs_link);
3382EXPORT_SYMBOL(vfs_mkdir);
3383EXPORT_SYMBOL(vfs_mknod);
3384EXPORT_SYMBOL(generic_permission);
3385EXPORT_SYMBOL(vfs_readlink);
3386EXPORT_SYMBOL(vfs_rename);
3387EXPORT_SYMBOL(vfs_rmdir);
3388EXPORT_SYMBOL(vfs_symlink);
3389EXPORT_SYMBOL(vfs_unlink);
3390EXPORT_SYMBOL(dentry_unhash);
3391EXPORT_SYMBOL(generic_readlink);