blob: 2826db35dc257bb15241376e3b6611582b2e4c92 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070035#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37
Eric Parise81e3f42009-12-04 15:47:36 -050038#include "internal.h"
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030074 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
Al Virof52e0c12011-03-14 18:56:51 -0400140static char *getname_flags(const char __user * filename, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
Al Virof52e0c12011-03-14 18:56:51 -0400151 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
152 __putname(tmp);
153 result = ERR_PTR(retval);
154 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
156 }
157 audit_getname(result);
158 return result;
159}
160
Al Virof52e0c12011-03-14 18:56:51 -0400161char *getname(const char __user * filename)
162{
163 return getname_flags(filename, 0);
164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166#ifdef CONFIG_AUDITSYSCALL
167void putname(const char *name)
168{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400169 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 audit_putname(name);
171 else
172 __putname(name);
173}
174EXPORT_SYMBOL(putname);
175#endif
176
Linus Torvaldse77819e2011-07-22 19:30:19 -0700177static int check_acl(struct inode *inode, int mask)
178{
Linus Torvalds84635d62011-07-25 22:47:03 -0700179#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700180 struct posix_acl *acl;
181
Linus Torvaldse77819e2011-07-22 19:30:19 -0700182 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400183 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
184 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700185 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400186 /* no ->get_acl() calls in RCU mode... */
187 if (acl == ACL_NOT_CACHED)
188 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300189 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700190 }
191
192 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
193
194 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200195 * A filesystem can force a ACL callback by just never filling the
196 * ACL cache. But normally you'd fill the cache either at inode
197 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700198 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200199 * If the filesystem doesn't have a get_acl() function at all, we'll
200 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700201 */
202 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200203 if (inode->i_op->get_acl) {
204 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
205 if (IS_ERR(acl))
206 return PTR_ERR(acl);
207 } else {
208 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
209 return -EAGAIN;
210 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700211 }
212
213 if (acl) {
214 int error = posix_acl_permission(inode, acl, mask);
215 posix_acl_release(acl);
216 return error;
217 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700218#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700219
220 return -EAGAIN;
221}
222
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700223/*
224 * This does basic POSIX ACL permission checking
225 */
Al Viro7e401452011-06-20 19:12:17 -0400226static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700227{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700228 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229
Al Viro9c2c7032011-06-20 19:06:22 -0400230 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700231
Serge E. Hallyne795b712011-03-23 16:43:25 -0700232 if (current_user_ns() != inode_userns(inode))
233 goto other_perms;
234
Linus Torvalds14067ff2011-07-25 19:55:52 -0700235 if (likely(current_fsuid() == inode->i_uid))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700236 mode >>= 6;
237 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400239 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100240 if (error != -EAGAIN)
241 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700242 }
243
244 if (in_group_p(inode->i_gid))
245 mode >>= 3;
246 }
247
Serge E. Hallyne795b712011-03-23 16:43:25 -0700248other_perms:
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700249 /*
250 * If the DACs are ok we don't need any capability check.
251 */
Al Viro9c2c7032011-06-20 19:06:22 -0400252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700253 return 0;
254 return -EACCES;
255}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100258 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * @inode: inode to check access rights for
260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100265 * are used for other things.
266 *
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 */
Al Viro2830ba72011-06-20 19:16:29 -0400271int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700273 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700276 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 */
Al Viro7e401452011-06-20 19:12:17 -0400278 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700279 if (ret != -EACCES)
280 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Al Virod594e7e2011-06-20 19:55:42 -0400282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
285 return 0;
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
288 return 0;
289 return -EACCES;
290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 /*
292 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 */
Al Virod594e7e2011-06-20 19:55:42 -0400296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 return 0;
299
300 /*
301 * Searching includes executable on directories, else just read.
302 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400304 if (mask == MAY_READ)
Serge E. Hallyne795b712011-03-23 16:43:25 -0700305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return 0;
307
308 return -EACCES;
309}
310
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700311/*
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
316 */
317static inline int do_inode_permission(struct inode *inode, int mask)
318{
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
322
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
327 }
328 return generic_permission(inode, mask);
329}
330
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200331/**
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
335 *
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
340 */
Al Virof419a2e2008-07-22 00:07:17 -0400341int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Al Viroe6305c42008-07-15 21:03:57 -0400343 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700345 if (unlikely(mask & MAY_WRITE)) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700346 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 /*
349 * Nobody gets write access to a read-only fs.
350 */
351 if (IS_RDONLY(inode) &&
352 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
353 return -EROFS;
354
355 /*
356 * Nobody gets write access to an immutable file.
357 */
358 if (IS_IMMUTABLE(inode))
359 return -EACCES;
360 }
361
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700362 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (retval)
364 return retval;
365
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700366 retval = devcgroup_inode_permission(inode, mask);
367 if (retval)
368 return retval;
369
Eric Parisd09ca732010-07-23 11:43:57 -0400370 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Al Virof4d6ff82011-06-19 13:14:21 -0400373/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800374 * path_get - get a reference to a path
375 * @path: path to get the reference to
376 *
377 * Given a path increment the reference count to the dentry and the vfsmount.
378 */
379void path_get(struct path *path)
380{
381 mntget(path->mnt);
382 dget(path->dentry);
383}
384EXPORT_SYMBOL(path_get);
385
386/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800387 * path_put - put a reference to a path
388 * @path: path to put the reference to
389 *
390 * Given a path decrement the reference count to the dentry and the vfsmount.
391 */
392void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Jan Blunck1d957f92008-02-14 19:34:35 -0800394 dput(path->dentry);
395 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
Jan Blunck1d957f92008-02-14 19:34:35 -0800397EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Al Viro19660af2011-03-25 10:32:48 -0400399/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100400 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400401 * Documentation/filesystems/path-lookup.txt). In situations when we can't
402 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
403 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
404 * mode. Refcounts are grabbed at the last known good point before rcu-walk
405 * got stuck, so ref-walk may continue from there. If this is not successful
406 * (eg. a seqcount has changed), then failure is returned and it's up to caller
407 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100408 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100409
410/**
Al Viro19660af2011-03-25 10:32:48 -0400411 * unlazy_walk - try to switch to ref-walk mode.
412 * @nd: nameidata pathwalk data
413 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800414 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100415 *
Al Viro19660af2011-03-25 10:32:48 -0400416 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
417 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
418 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100419 */
Al Viro19660af2011-03-25 10:32:48 -0400420static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100421{
422 struct fs_struct *fs = current->fs;
423 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500424 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100425
426 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500427 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
428 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100429 spin_lock(&fs->lock);
430 if (nd->root.mnt != fs->root.mnt ||
431 nd->root.dentry != fs->root.dentry)
432 goto err_root;
433 }
434 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400435 if (!dentry) {
436 if (!__d_rcu_to_refcount(parent, nd->seq))
437 goto err_parent;
438 BUG_ON(nd->inode != parent->d_inode);
439 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400440 if (dentry->d_parent != parent)
441 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400442 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
443 if (!__d_rcu_to_refcount(dentry, nd->seq))
444 goto err_child;
445 /*
446 * If the sequence check on the child dentry passed, then
447 * the child has not been removed from its parent. This
448 * means the parent dentry must be valid and able to take
449 * a reference at this point.
450 */
451 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
452 BUG_ON(!parent->d_count);
453 parent->d_count++;
454 spin_unlock(&dentry->d_lock);
455 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100456 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500457 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100458 path_get(&nd->root);
459 spin_unlock(&fs->lock);
460 }
461 mntget(nd->path.mnt);
462
463 rcu_read_unlock();
464 br_read_unlock(vfsmount_lock);
465 nd->flags &= ~LOOKUP_RCU;
466 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400467
468err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100469 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400470err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100471 spin_unlock(&parent->d_lock);
472err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500473 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100474 spin_unlock(&fs->lock);
475 return -ECHILD;
476}
477
Nick Piggin31e6b012011-01-07 17:49:52 +1100478/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700479 * release_open_intent - free up open intent resources
480 * @nd: pointer to nameidata
481 */
482void release_open_intent(struct nameidata *nd)
483{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800484 struct file *file = nd->intent.open.file;
485
486 if (file && !IS_ERR(file)) {
487 if (file->f_path.dentry == NULL)
488 put_filp(file);
489 else
490 fput(file);
491 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700492}
493
Al Virof60aef72011-02-15 01:35:28 -0500494static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100495{
Al Virof60aef72011-02-15 01:35:28 -0500496 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100497}
498
Al Viro9f1fafe2011-03-25 11:00:12 -0400499/**
500 * complete_walk - successful completion of path walk
501 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500502 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400503 * If we had been in RCU mode, drop out of it and legitimize nd->path.
504 * Revalidate the final result, unless we'd already done that during
505 * the path walk or the filesystem doesn't ask for it. Return 0 on
506 * success, -error on failure. In case of failure caller does not
507 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500508 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400509static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500510{
Al Viro16c2cd72011-02-22 15:50:10 -0500511 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500512 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500513
Al Viro9f1fafe2011-03-25 11:00:12 -0400514 if (nd->flags & LOOKUP_RCU) {
515 nd->flags &= ~LOOKUP_RCU;
516 if (!(nd->flags & LOOKUP_ROOT))
517 nd->root.mnt = NULL;
518 spin_lock(&dentry->d_lock);
519 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
520 spin_unlock(&dentry->d_lock);
521 rcu_read_unlock();
522 br_read_unlock(vfsmount_lock);
523 return -ECHILD;
524 }
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
527 mntget(nd->path.mnt);
528 rcu_read_unlock();
529 br_read_unlock(vfsmount_lock);
530 }
531
Al Viro16c2cd72011-02-22 15:50:10 -0500532 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500533 return 0;
534
Al Viro16c2cd72011-02-22 15:50:10 -0500535 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
536 return 0;
537
538 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
539 return 0;
540
541 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100542 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500543 if (status > 0)
544 return 0;
545
Al Viro16c2cd72011-02-22 15:50:10 -0500546 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500547 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500548
Al Viro9f1fafe2011-03-25 11:00:12 -0400549 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500550 return status;
551}
552
Al Viro2a737872009-04-07 11:49:53 -0400553static __always_inline void set_root(struct nameidata *nd)
554{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200555 if (!nd->root.mnt)
556 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400557}
558
Al Viro6de88d72009-08-09 01:41:57 +0400559static int link_path_walk(const char *, struct nameidata *);
560
Nick Piggin31e6b012011-01-07 17:49:52 +1100561static __always_inline void set_root_rcu(struct nameidata *nd)
562{
563 if (!nd->root.mnt) {
564 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100565 unsigned seq;
566
567 do {
568 seq = read_seqcount_begin(&fs->seq);
569 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700570 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100571 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100572 }
573}
574
Arjan van de Venf1662352006-01-14 13:21:31 -0800575static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Nick Piggin31e6b012011-01-07 17:49:52 +1100577 int ret;
578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (IS_ERR(link))
580 goto fail;
581
582 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400583 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800584 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400585 nd->path = nd->root;
586 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500587 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100589 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100590
Nick Piggin31e6b012011-01-07 17:49:52 +1100591 ret = link_path_walk(link, nd);
592 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800594 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 return PTR_ERR(link);
596}
597
Jan Blunck1d957f92008-02-14 19:34:35 -0800598static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700599{
600 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800601 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700602 mntput(path->mnt);
603}
604
Nick Piggin7b9337a2011-01-14 08:42:43 +0000605static inline void path_to_nameidata(const struct path *path,
606 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700607{
Nick Piggin31e6b012011-01-07 17:49:52 +1100608 if (!(nd->flags & LOOKUP_RCU)) {
609 dput(nd->path.dentry);
610 if (nd->path.mnt != path->mnt)
611 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800612 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100613 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800614 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700615}
616
Al Viro574197e2011-03-14 22:20:34 -0400617static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
618{
619 struct inode *inode = link->dentry->d_inode;
620 if (!IS_ERR(cookie) && inode->i_op->put_link)
621 inode->i_op->put_link(link->dentry, nd, cookie);
622 path_put(link);
623}
624
Al Virodef4af32009-12-26 08:37:05 -0500625static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400626follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800627{
628 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000629 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800630
Al Viro844a3912011-02-15 00:38:26 -0500631 BUG_ON(nd->flags & LOOKUP_RCU);
632
Al Viro0e794582011-03-16 02:45:02 -0400633 if (link->mnt == nd->path.mnt)
634 mntget(link->mnt);
635
Al Viro574197e2011-03-14 22:20:34 -0400636 if (unlikely(current->total_link_count >= 40)) {
637 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400638 path_put(&nd->path);
639 return -ELOOP;
640 }
641 cond_resched();
642 current->total_link_count++;
643
Nick Piggin7b9337a2011-01-14 08:42:43 +0000644 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800645 nd_set_link(nd, NULL);
646
Al Viro36f3b4f2011-02-22 21:24:38 -0500647 error = security_inode_follow_link(link->dentry, nd);
648 if (error) {
649 *p = ERR_PTR(error); /* no ->put_link(), please */
650 path_put(&nd->path);
651 return error;
652 }
653
Al Viro86acdca12009-12-22 23:45:11 -0500654 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500655 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
656 error = PTR_ERR(*p);
657 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800658 char *s = nd_get_link(nd);
659 error = 0;
660 if (s)
661 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400662 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500663 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400664 nd->inode = nd->path.dentry->d_inode;
665 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400666 /* stepped on a _really_ weird one */
667 path_put(&nd->path);
668 error = -ELOOP;
669 }
670 }
Ian Kent051d3812006-03-27 01:14:53 -0800671 }
Ian Kent051d3812006-03-27 01:14:53 -0800672 return error;
673}
674
Nick Piggin31e6b012011-01-07 17:49:52 +1100675static int follow_up_rcu(struct path *path)
676{
677 struct vfsmount *parent;
678 struct dentry *mountpoint;
679
680 parent = path->mnt->mnt_parent;
681 if (parent == path->mnt)
682 return 0;
683 mountpoint = path->mnt->mnt_mountpoint;
684 path->dentry = mountpoint;
685 path->mnt = parent;
686 return 1;
687}
688
Al Virobab77eb2009-04-18 03:26:48 -0400689int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 struct vfsmount *parent;
692 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000693
694 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400695 parent = path->mnt->mnt_parent;
696 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000697 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 return 0;
699 }
700 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400701 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000702 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400703 dput(path->dentry);
704 path->dentry = mountpoint;
705 mntput(path->mnt);
706 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 return 1;
708}
709
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100710/*
David Howells9875cf82011-01-14 18:45:21 +0000711 * Perform an automount
712 * - return -EISDIR to tell follow_managed() to stop and return the path we
713 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 */
David Howells9875cf82011-01-14 18:45:21 +0000715static int follow_automount(struct path *path, unsigned flags,
716 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100717{
David Howells9875cf82011-01-14 18:45:21 +0000718 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000719 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100720
David Howells9875cf82011-01-14 18:45:21 +0000721 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
722 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700723
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
David Howells5a30d8a2011-07-11 14:20:57 +0100730 /*
David Howells9875cf82011-01-14 18:45:21 +0000731 * We don't want to mount if someone's just doing a stat and they've
732 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
733 * appended a '/' to the name.
734 */
David Howells5a30d8a2011-07-11 14:20:57 +0100735 if (!(flags & LOOKUP_FOLLOW)) {
736 /* We do, however, want to mount if someone wants to open or
737 * create a file of any type under the mountpoint, wants to
738 * traverse through the mountpoint or wants to open the mounted
739 * directory.
740 * Also, autofs may mark negative dentries as being automount
741 * points. These will need the attentions of the daemon to
742 * instantiate them before they can be used.
743 */
744 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
745 LOOKUP_OPEN | LOOKUP_CREATE)) &&
746 path->dentry->d_inode)
747 return -EISDIR;
748 }
David Howells9875cf82011-01-14 18:45:21 +0000749 current->total_link_count++;
750 if (current->total_link_count >= 40)
751 return -ELOOP;
752
753 mnt = path->dentry->d_op->d_automount(path);
754 if (IS_ERR(mnt)) {
755 /*
756 * The filesystem is allowed to return -EISDIR here to indicate
757 * it doesn't want to automount. For instance, autofs would do
758 * this so that its userspace daemon can mount on this dentry.
759 *
760 * However, we can only permit this if it's a terminal point in
761 * the path being looked up; if it wasn't then the remainder of
762 * the path is inaccessible and we should say so.
763 */
Al Viro49084c32011-06-25 21:59:52 -0400764 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000765 return -EREMOTE;
766 return PTR_ERR(mnt);
767 }
David Howellsea5b7782011-01-14 19:10:03 +0000768
David Howells9875cf82011-01-14 18:45:21 +0000769 if (!mnt) /* mount collision */
770 return 0;
771
Al Viro8aef1882011-06-16 15:10:06 +0100772 if (!*need_mntput) {
773 /* lock_mount() may release path->mnt on error */
774 mntget(path->mnt);
775 *need_mntput = true;
776 }
Al Viro19a167a2011-01-17 01:35:23 -0500777 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000778
David Howellsea5b7782011-01-14 19:10:03 +0000779 switch (err) {
780 case -EBUSY:
781 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500782 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000783 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100784 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000785 path->mnt = mnt;
786 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000787 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500788 default:
789 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000790 }
Al Viro19a167a2011-01-17 01:35:23 -0500791
David Howells9875cf82011-01-14 18:45:21 +0000792}
793
794/*
795 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000796 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000797 * - Flagged as mountpoint
798 * - Flagged as automount point
799 *
800 * This may only be called in refwalk mode.
801 *
802 * Serialization is taken care of in namespace.c
803 */
804static int follow_managed(struct path *path, unsigned flags)
805{
Al Viro8aef1882011-06-16 15:10:06 +0100806 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000807 unsigned managed;
808 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100809 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000810
811 /* Given that we're not holding a lock here, we retain the value in a
812 * local variable for each dentry as we look at it so that we don't see
813 * the components of that value change under us */
814 while (managed = ACCESS_ONCE(path->dentry->d_flags),
815 managed &= DCACHE_MANAGED_DENTRY,
816 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000817 /* Allow the filesystem to manage the transit without i_mutex
818 * being held. */
819 if (managed & DCACHE_MANAGE_TRANSIT) {
820 BUG_ON(!path->dentry->d_op);
821 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400822 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000823 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100824 break;
David Howellscc53ce52011-01-14 18:45:26 +0000825 }
826
David Howells9875cf82011-01-14 18:45:21 +0000827 /* Transit to a mounted filesystem. */
828 if (managed & DCACHE_MOUNTED) {
829 struct vfsmount *mounted = lookup_mnt(path);
830 if (mounted) {
831 dput(path->dentry);
832 if (need_mntput)
833 mntput(path->mnt);
834 path->mnt = mounted;
835 path->dentry = dget(mounted->mnt_root);
836 need_mntput = true;
837 continue;
838 }
839
840 /* Something is mounted on this dentry in another
841 * namespace and/or whatever was mounted there in this
842 * namespace got unmounted before we managed to get the
843 * vfsmount_lock */
844 }
845
846 /* Handle an automount point */
847 if (managed & DCACHE_NEED_AUTOMOUNT) {
848 ret = follow_automount(path, flags, &need_mntput);
849 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100850 break;
David Howells9875cf82011-01-14 18:45:21 +0000851 continue;
852 }
853
854 /* We didn't change the current path point */
855 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 }
Al Viro8aef1882011-06-16 15:10:06 +0100857
858 if (need_mntput && path->mnt == mnt)
859 mntput(path->mnt);
860 if (ret == -EISDIR)
861 ret = 0;
862 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
864
David Howellscc53ce52011-01-14 18:45:26 +0000865int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
867 struct vfsmount *mounted;
868
Al Viro1c755af2009-04-18 14:06:57 -0400869 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400871 dput(path->dentry);
872 mntput(path->mnt);
873 path->mnt = mounted;
874 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 return 1;
876 }
877 return 0;
878}
879
Ian Kent62a73752011-03-25 01:51:02 +0800880static inline bool managed_dentry_might_block(struct dentry *dentry)
881{
882 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
883 dentry->d_op->d_manage(dentry, true) < 0);
884}
885
David Howells9875cf82011-01-14 18:45:21 +0000886/*
Al Viro287548e2011-05-27 06:50:06 -0400887 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
888 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000889 */
890static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400891 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000892{
Ian Kent62a73752011-03-25 01:51:02 +0800893 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +0000894 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800895 /*
896 * Don't forget we might have a non-mountpoint managed dentry
897 * that wants to block transit.
898 */
Al Viro287548e2011-05-27 06:50:06 -0400899 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000900 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800901
902 if (!d_mountpoint(path->dentry))
903 break;
904
David Howells9875cf82011-01-14 18:45:21 +0000905 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
906 if (!mounted)
907 break;
908 path->mnt = mounted;
909 path->dentry = mounted->mnt_root;
910 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700911 /*
912 * Update the inode too. We don't need to re-check the
913 * dentry sequence number here after this d_inode read,
914 * because a mount-point is always pinned.
915 */
916 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000917 }
David Howells9875cf82011-01-14 18:45:21 +0000918 return true;
919}
920
Al Virodea39372011-05-27 06:53:39 -0400921static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400922{
Al Virodea39372011-05-27 06:53:39 -0400923 while (d_mountpoint(nd->path.dentry)) {
Al Viro287548e2011-05-27 06:50:06 -0400924 struct vfsmount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400925 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400926 if (!mounted)
927 break;
Al Virodea39372011-05-27 06:53:39 -0400928 nd->path.mnt = mounted;
929 nd->path.dentry = mounted->mnt_root;
930 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400931 }
932}
933
Nick Piggin31e6b012011-01-07 17:49:52 +1100934static int follow_dotdot_rcu(struct nameidata *nd)
935{
Nick Piggin31e6b012011-01-07 17:49:52 +1100936 set_root_rcu(nd);
937
David Howells9875cf82011-01-14 18:45:21 +0000938 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100939 if (nd->path.dentry == nd->root.dentry &&
940 nd->path.mnt == nd->root.mnt) {
941 break;
942 }
943 if (nd->path.dentry != nd->path.mnt->mnt_root) {
944 struct dentry *old = nd->path.dentry;
945 struct dentry *parent = old->d_parent;
946 unsigned seq;
947
948 seq = read_seqcount_begin(&parent->d_seq);
949 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500950 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100951 nd->path.dentry = parent;
952 nd->seq = seq;
953 break;
954 }
955 if (!follow_up_rcu(&nd->path))
956 break;
957 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100958 }
Al Virodea39372011-05-27 06:53:39 -0400959 follow_mount_rcu(nd);
960 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100961 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500962
963failed:
964 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500965 if (!(nd->flags & LOOKUP_ROOT))
966 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500967 rcu_read_unlock();
968 br_read_unlock(vfsmount_lock);
969 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100970}
971
David Howells9875cf82011-01-14 18:45:21 +0000972/*
David Howellscc53ce52011-01-14 18:45:26 +0000973 * Follow down to the covering mount currently visible to userspace. At each
974 * point, the filesystem owning that dentry may be queried as to whether the
975 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000976 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400977int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000978{
979 unsigned managed;
980 int ret;
981
982 while (managed = ACCESS_ONCE(path->dentry->d_flags),
983 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
984 /* Allow the filesystem to manage the transit without i_mutex
985 * being held.
986 *
987 * We indicate to the filesystem if someone is trying to mount
988 * something here. This gives autofs the chance to deny anyone
989 * other than its daemon the right to mount on its
990 * superstructure.
991 *
992 * The filesystem may sleep at this point.
993 */
994 if (managed & DCACHE_MANAGE_TRANSIT) {
995 BUG_ON(!path->dentry->d_op);
996 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000997 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400998 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000999 if (ret < 0)
1000 return ret == -EISDIR ? 0 : ret;
1001 }
1002
1003 /* Transit to a mounted filesystem. */
1004 if (managed & DCACHE_MOUNTED) {
1005 struct vfsmount *mounted = lookup_mnt(path);
1006 if (!mounted)
1007 break;
1008 dput(path->dentry);
1009 mntput(path->mnt);
1010 path->mnt = mounted;
1011 path->dentry = dget(mounted->mnt_root);
1012 continue;
1013 }
1014
1015 /* Don't handle automount points here */
1016 break;
1017 }
1018 return 0;
1019}
1020
1021/*
David Howells9875cf82011-01-14 18:45:21 +00001022 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1023 */
1024static void follow_mount(struct path *path)
1025{
1026 while (d_mountpoint(path->dentry)) {
1027 struct vfsmount *mounted = lookup_mnt(path);
1028 if (!mounted)
1029 break;
1030 dput(path->dentry);
1031 mntput(path->mnt);
1032 path->mnt = mounted;
1033 path->dentry = dget(mounted->mnt_root);
1034 }
1035}
1036
Nick Piggin31e6b012011-01-07 17:49:52 +11001037static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038{
Al Viro2a737872009-04-07 11:49:53 -04001039 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001042 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Al Viro2a737872009-04-07 11:49:53 -04001044 if (nd->path.dentry == nd->root.dentry &&
1045 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 break;
1047 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001048 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001049 /* rare case of legitimate dget_parent()... */
1050 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 dput(old);
1052 break;
1053 }
Al Viro3088dd72010-01-30 15:47:29 -05001054 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 }
Al Viro79ed0222009-04-18 13:59:41 -04001057 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001058 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059}
1060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001062 * Allocate a dentry with name and parent, and perform a parent
1063 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1064 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1065 * have verified that no child exists while under i_mutex.
1066 */
1067static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1068 struct qstr *name, struct nameidata *nd)
1069{
1070 struct inode *inode = parent->d_inode;
1071 struct dentry *dentry;
1072 struct dentry *old;
1073
1074 /* Don't create child dentry for a dead directory. */
1075 if (unlikely(IS_DEADDIR(inode)))
1076 return ERR_PTR(-ENOENT);
1077
1078 dentry = d_alloc(parent, name);
1079 if (unlikely(!dentry))
1080 return ERR_PTR(-ENOMEM);
1081
1082 old = inode->i_op->lookup(inode, dentry, nd);
1083 if (unlikely(old)) {
1084 dput(dentry);
1085 dentry = old;
1086 }
1087 return dentry;
1088}
1089
1090/*
Josef Bacik44396f42011-05-31 11:58:49 -04001091 * We already have a dentry, but require a lookup to be performed on the parent
1092 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1093 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1094 * child exists while under i_mutex.
1095 */
1096static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1097 struct nameidata *nd)
1098{
1099 struct inode *inode = parent->d_inode;
1100 struct dentry *old;
1101
1102 /* Don't create child dentry for a dead directory. */
1103 if (unlikely(IS_DEADDIR(inode)))
1104 return ERR_PTR(-ENOENT);
1105
1106 old = inode->i_op->lookup(inode, dentry, nd);
1107 if (unlikely(old)) {
1108 dput(dentry);
1109 dentry = old;
1110 }
1111 return dentry;
1112}
1113
1114/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 * It's more convoluted than I'd like it to be, but... it's still fairly
1116 * small and for now I'd prefer to have fast path as straight as possible.
1117 * It _is_ time-critical.
1118 */
1119static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001120 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
Jan Blunck4ac91372008-02-14 19:34:32 -08001122 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001123 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001124 int need_reval = 1;
1125 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001126 int err;
1127
Al Viro3cac2602009-08-13 18:27:43 +04001128 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001129 * Rename seqlock is not required here because in the off chance
1130 * of a false negative due to a concurrent rename, we're going to
1131 * do the non-racy lookup, below.
1132 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001133 if (nd->flags & LOOKUP_RCU) {
1134 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001135 *inode = nd->inode;
1136 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001137 if (!dentry)
1138 goto unlazy;
1139
Nick Piggin31e6b012011-01-07 17:49:52 +11001140 /* Memory barrier in read_seqcount_begin of child is enough */
1141 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1142 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001143 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001144
Al Viro24643082011-02-15 01:26:22 -05001145 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001146 status = d_revalidate(dentry, nd);
1147 if (unlikely(status <= 0)) {
1148 if (status != -ECHILD)
1149 need_reval = 0;
1150 goto unlazy;
1151 }
Al Viro24643082011-02-15 01:26:22 -05001152 }
Josef Bacik44396f42011-05-31 11:58:49 -04001153 if (unlikely(d_need_lookup(dentry)))
1154 goto unlazy;
Nick Piggin31e6b012011-01-07 17:49:52 +11001155 path->mnt = mnt;
1156 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001157 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1158 goto unlazy;
1159 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1160 goto unlazy;
1161 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001162unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001163 if (unlazy_walk(nd, dentry))
1164 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001165 } else {
1166 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001167 }
Al Viro5a18fff2011-03-11 04:44:53 -05001168
Josef Bacik44396f42011-05-31 11:58:49 -04001169 if (dentry && unlikely(d_need_lookup(dentry))) {
1170 dput(dentry);
1171 dentry = NULL;
1172 }
Al Viro5a18fff2011-03-11 04:44:53 -05001173retry:
1174 if (unlikely(!dentry)) {
1175 struct inode *dir = parent->d_inode;
1176 BUG_ON(nd->inode != dir);
1177
1178 mutex_lock(&dir->i_mutex);
1179 dentry = d_lookup(parent, name);
1180 if (likely(!dentry)) {
1181 dentry = d_alloc_and_lookup(parent, name, nd);
1182 if (IS_ERR(dentry)) {
1183 mutex_unlock(&dir->i_mutex);
1184 return PTR_ERR(dentry);
1185 }
1186 /* known good */
1187 need_reval = 0;
1188 status = 1;
Josef Bacik44396f42011-05-31 11:58:49 -04001189 } else if (unlikely(d_need_lookup(dentry))) {
1190 dentry = d_inode_lookup(parent, dentry, nd);
1191 if (IS_ERR(dentry)) {
1192 mutex_unlock(&dir->i_mutex);
1193 return PTR_ERR(dentry);
1194 }
1195 /* known good */
1196 need_reval = 0;
1197 status = 1;
Al Viro5a18fff2011-03-11 04:44:53 -05001198 }
1199 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001200 }
Al Viro5a18fff2011-03-11 04:44:53 -05001201 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1202 status = d_revalidate(dentry, nd);
1203 if (unlikely(status <= 0)) {
1204 if (status < 0) {
1205 dput(dentry);
1206 return status;
1207 }
1208 if (!d_invalidate(dentry)) {
1209 dput(dentry);
1210 dentry = NULL;
1211 need_reval = 1;
1212 goto retry;
1213 }
1214 }
1215
David Howells9875cf82011-01-14 18:45:21 +00001216 path->mnt = mnt;
1217 path->dentry = dentry;
1218 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001219 if (unlikely(err < 0)) {
1220 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001221 return err;
Ian Kent89312212011-01-18 12:06:10 +08001222 }
David Howells9875cf82011-01-14 18:45:21 +00001223 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225}
1226
Al Viro52094c82011-02-21 21:34:47 -05001227static inline int may_lookup(struct nameidata *nd)
1228{
1229 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001230 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001231 if (err != -ECHILD)
1232 return err;
Al Viro19660af2011-03-25 10:32:48 -04001233 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001234 return -ECHILD;
1235 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001236 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001237}
1238
Al Viro9856fa12011-03-04 14:22:06 -05001239static inline int handle_dots(struct nameidata *nd, int type)
1240{
1241 if (type == LAST_DOTDOT) {
1242 if (nd->flags & LOOKUP_RCU) {
1243 if (follow_dotdot_rcu(nd))
1244 return -ECHILD;
1245 } else
1246 follow_dotdot(nd);
1247 }
1248 return 0;
1249}
1250
Al Viro951361f2011-03-04 14:44:37 -05001251static void terminate_walk(struct nameidata *nd)
1252{
1253 if (!(nd->flags & LOOKUP_RCU)) {
1254 path_put(&nd->path);
1255 } else {
1256 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001257 if (!(nd->flags & LOOKUP_ROOT))
1258 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001259 rcu_read_unlock();
1260 br_read_unlock(vfsmount_lock);
1261 }
1262}
1263
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001264/*
1265 * Do we need to follow links? We _really_ want to be able
1266 * to do this check without having to look at inode->i_op,
1267 * so we keep a cache of "no, this doesn't need follow_link"
1268 * for the common case.
1269 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001270static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001271{
1272 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1273 if (likely(inode->i_op->follow_link))
1274 return follow;
1275
1276 /* This gets set once for the inode lifetime */
1277 spin_lock(&inode->i_lock);
1278 inode->i_opflags |= IOP_NOFOLLOW;
1279 spin_unlock(&inode->i_lock);
1280 }
1281 return 0;
1282}
1283
Al Viroce57dfc2011-03-13 19:58:58 -04001284static inline int walk_component(struct nameidata *nd, struct path *path,
1285 struct qstr *name, int type, int follow)
1286{
1287 struct inode *inode;
1288 int err;
1289 /*
1290 * "." and ".." are special - ".." especially so because it has
1291 * to be able to know about the current root directory and
1292 * parent relationships.
1293 */
1294 if (unlikely(type != LAST_NORM))
1295 return handle_dots(nd, type);
1296 err = do_lookup(nd, name, path, &inode);
1297 if (unlikely(err)) {
1298 terminate_walk(nd);
1299 return err;
1300 }
1301 if (!inode) {
1302 path_to_nameidata(path, nd);
1303 terminate_walk(nd);
1304 return -ENOENT;
1305 }
Linus Torvalds7813b942011-08-07 09:53:20 -07001306 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001307 if (nd->flags & LOOKUP_RCU) {
1308 if (unlikely(unlazy_walk(nd, path->dentry))) {
1309 terminate_walk(nd);
1310 return -ECHILD;
1311 }
1312 }
Al Viroce57dfc2011-03-13 19:58:58 -04001313 BUG_ON(inode != path->dentry->d_inode);
1314 return 1;
1315 }
1316 path_to_nameidata(path, nd);
1317 nd->inode = inode;
1318 return 0;
1319}
1320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321/*
Al Virob3563792011-03-14 21:54:55 -04001322 * This limits recursive symlink follows to 8, while
1323 * limiting consecutive symlinks to 40.
1324 *
1325 * Without that kind of total limit, nasty chains of consecutive
1326 * symlinks can cause almost arbitrarily long lookups.
1327 */
1328static inline int nested_symlink(struct path *path, struct nameidata *nd)
1329{
1330 int res;
1331
Al Virob3563792011-03-14 21:54:55 -04001332 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1333 path_put_conditional(path, nd);
1334 path_put(&nd->path);
1335 return -ELOOP;
1336 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001337 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001338
1339 nd->depth++;
1340 current->link_count++;
1341
1342 do {
1343 struct path link = *path;
1344 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001345
1346 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001347 if (!res)
1348 res = walk_component(nd, path, &nd->last,
1349 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001350 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001351 } while (res > 0);
1352
1353 current->link_count--;
1354 nd->depth--;
1355 return res;
1356}
1357
1358/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001359 * We really don't want to look at inode->i_op->lookup
1360 * when we don't have to. So we keep a cache bit in
1361 * the inode ->i_opflags field that says "yes, we can
1362 * do lookup on this inode".
1363 */
1364static inline int can_lookup(struct inode *inode)
1365{
1366 if (likely(inode->i_opflags & IOP_LOOKUP))
1367 return 1;
1368 if (likely(!inode->i_op->lookup))
1369 return 0;
1370
1371 /* We do this once for the lifetime of the inode */
1372 spin_lock(&inode->i_lock);
1373 inode->i_opflags |= IOP_LOOKUP;
1374 spin_unlock(&inode->i_lock);
1375 return 1;
1376}
1377
1378/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001380 * This is the basic name resolution function, turning a pathname into
1381 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001383 * Returns 0 and nd will have valid dentry and mnt on success.
1384 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 */
Al Viro6de88d72009-08-09 01:41:57 +04001386static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387{
1388 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 while (*name=='/')
1392 name++;
1393 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001394 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 /* At this point we know we have a real path component. */
1397 for(;;) {
1398 unsigned long hash;
1399 struct qstr this;
1400 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001401 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
Al Viro52094c82011-02-21 21:34:47 -05001403 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (err)
1405 break;
1406
1407 this.name = name;
1408 c = *(const unsigned char *)name;
1409
1410 hash = init_name_hash();
1411 do {
1412 name++;
1413 hash = partial_name_hash(c, hash);
1414 c = *(const unsigned char *)name;
1415 } while (c && (c != '/'));
1416 this.len = name - (const char *) this.name;
1417 this.hash = end_name_hash(hash);
1418
Al Virofe479a52011-02-22 15:10:03 -05001419 type = LAST_NORM;
1420 if (this.name[0] == '.') switch (this.len) {
1421 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001422 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001423 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001424 nd->flags |= LOOKUP_JUMPED;
1425 }
Al Virofe479a52011-02-22 15:10:03 -05001426 break;
1427 case 1:
1428 type = LAST_DOT;
1429 }
Al Viro5a202bc2011-03-08 14:17:44 -05001430 if (likely(type == LAST_NORM)) {
1431 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001432 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001433 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1434 err = parent->d_op->d_hash(parent, nd->inode,
1435 &this);
1436 if (err < 0)
1437 break;
1438 }
1439 }
Al Virofe479a52011-02-22 15:10:03 -05001440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 /* remove trailing slashes? */
1442 if (!c)
1443 goto last_component;
1444 while (*++name == '/');
1445 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001446 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Al Viroce57dfc2011-03-13 19:58:58 -04001448 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1449 if (err < 0)
1450 return err;
Al Virofe479a52011-02-22 15:10:03 -05001451
Al Viroce57dfc2011-03-13 19:58:58 -04001452 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001453 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001455 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001456 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001457 if (can_lookup(nd->inode))
1458 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001460 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 /* here ends the main loop */
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463last_component:
Al Virob3563792011-03-14 21:54:55 -04001464 nd->last = this;
1465 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
Al Viro951361f2011-03-04 14:44:37 -05001468 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 return err;
1470}
1471
Al Viro70e9b352011-03-05 21:12:22 -05001472static int path_init(int dfd, const char *name, unsigned int flags,
1473 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001475 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001476 int fput_needed;
1477 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
1479 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001480 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001482 if (flags & LOOKUP_ROOT) {
1483 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001484 if (*name) {
1485 if (!inode->i_op->lookup)
1486 return -ENOTDIR;
1487 retval = inode_permission(inode, MAY_EXEC);
1488 if (retval)
1489 return retval;
1490 }
Al Viro5b6ca022011-03-09 23:04:47 -05001491 nd->path = nd->root;
1492 nd->inode = inode;
1493 if (flags & LOOKUP_RCU) {
1494 br_read_lock(vfsmount_lock);
1495 rcu_read_lock();
1496 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1497 } else {
1498 path_get(&nd->path);
1499 }
1500 return 0;
1501 }
1502
Al Viro2a737872009-04-07 11:49:53 -04001503 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001506 if (flags & LOOKUP_RCU) {
1507 br_read_lock(vfsmount_lock);
1508 rcu_read_lock();
1509 set_root_rcu(nd);
1510 } else {
1511 set_root(nd);
1512 path_get(&nd->root);
1513 }
Al Viro2a737872009-04-07 11:49:53 -04001514 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001515 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001516 if (flags & LOOKUP_RCU) {
1517 struct fs_struct *fs = current->fs;
1518 unsigned seq;
1519
1520 br_read_lock(vfsmount_lock);
1521 rcu_read_lock();
1522
1523 do {
1524 seq = read_seqcount_begin(&fs->seq);
1525 nd->path = fs->pwd;
1526 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1527 } while (read_seqcount_retry(&fs->seq, seq));
1528 } else {
1529 get_fs_pwd(current->fs, &nd->path);
1530 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001531 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001532 struct dentry *dentry;
1533
Al Viro1abf0c72011-03-13 03:51:11 -04001534 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001535 retval = -EBADF;
1536 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001537 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001538
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001539 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001540
Al Virof52e0c12011-03-14 18:56:51 -04001541 if (*name) {
1542 retval = -ENOTDIR;
1543 if (!S_ISDIR(dentry->d_inode->i_mode))
1544 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001545
Al Viro4ad5abb2011-06-20 19:57:03 -04001546 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001547 if (retval)
1548 goto fput_fail;
1549 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001550
Jan Blunck5dd784d02008-02-14 19:34:38 -08001551 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001552 if (flags & LOOKUP_RCU) {
1553 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001554 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001555 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1556 br_read_lock(vfsmount_lock);
1557 rcu_read_lock();
1558 } else {
1559 path_get(&file->f_path);
1560 fput_light(file, fput_needed);
1561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
Al Viroe41f7d42011-02-22 14:02:58 -05001563
Nick Piggin31e6b012011-01-07 17:49:52 +11001564 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001565 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001566
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001567fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001568 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001569out_fail:
1570 return retval;
1571}
1572
Al Virobd92d7f2011-03-14 19:54:59 -04001573static inline int lookup_last(struct nameidata *nd, struct path *path)
1574{
1575 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1576 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1577
1578 nd->flags &= ~LOOKUP_PARENT;
1579 return walk_component(nd, path, &nd->last, nd->last_type,
1580 nd->flags & LOOKUP_FOLLOW);
1581}
1582
Al Viro9b4a9b12009-04-07 11:44:16 -04001583/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001584static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001585 unsigned int flags, struct nameidata *nd)
1586{
Al Viro70e9b352011-03-05 21:12:22 -05001587 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001588 struct path path;
1589 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001590
1591 /*
1592 * Path walking is largely split up into 2 different synchronisation
1593 * schemes, rcu-walk and ref-walk (explained in
1594 * Documentation/filesystems/path-lookup.txt). These share much of the
1595 * path walk code, but some things particularly setup, cleanup, and
1596 * following mounts are sufficiently divergent that functions are
1597 * duplicated. Typically there is a function foo(), and its RCU
1598 * analogue, foo_rcu().
1599 *
1600 * -ECHILD is the error number of choice (just to avoid clashes) that
1601 * is returned if some aspect of an rcu-walk fails. Such an error must
1602 * be handled by restarting a traditional ref-walk (which will always
1603 * be able to complete).
1604 */
Al Virobd92d7f2011-03-14 19:54:59 -04001605 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001606
Al Virobd92d7f2011-03-14 19:54:59 -04001607 if (unlikely(err))
1608 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001609
1610 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001611 err = link_path_walk(name, nd);
1612
1613 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001614 err = lookup_last(nd, &path);
1615 while (err > 0) {
1616 void *cookie;
1617 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001618 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001619 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001620 if (!err)
1621 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001622 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001623 }
1624 }
Al Viroee0827c2011-02-21 23:38:09 -05001625
Al Viro9f1fafe2011-03-25 11:00:12 -04001626 if (!err)
1627 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001628
1629 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1630 if (!nd->inode->i_op->lookup) {
1631 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001632 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001633 }
1634 }
Al Viro16c2cd72011-02-22 15:50:10 -05001635
Al Viro70e9b352011-03-05 21:12:22 -05001636 if (base)
1637 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001638
Al Viro5b6ca022011-03-09 23:04:47 -05001639 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001640 path_put(&nd->root);
1641 nd->root.mnt = NULL;
1642 }
Al Virobd92d7f2011-03-14 19:54:59 -04001643 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001644}
Nick Piggin31e6b012011-01-07 17:49:52 +11001645
Al Viroee0827c2011-02-21 23:38:09 -05001646static int do_path_lookup(int dfd, const char *name,
1647 unsigned int flags, struct nameidata *nd)
1648{
1649 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1650 if (unlikely(retval == -ECHILD))
1651 retval = path_lookupat(dfd, name, flags, nd);
1652 if (unlikely(retval == -ESTALE))
1653 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001654
1655 if (likely(!retval)) {
1656 if (unlikely(!audit_dummy_context())) {
1657 if (nd->path.dentry && nd->inode)
1658 audit_inode(name, nd->path.dentry);
1659 }
1660 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001661 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662}
1663
Al Viroc9c6cac2011-02-16 15:15:47 -05001664int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001665{
Al Viroc9c6cac2011-02-16 15:15:47 -05001666 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001667}
1668
Al Virod1811462008-08-02 00:49:18 -04001669int kern_path(const char *name, unsigned int flags, struct path *path)
1670{
1671 struct nameidata nd;
1672 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1673 if (!res)
1674 *path = nd.path;
1675 return res;
1676}
1677
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001678/**
1679 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1680 * @dentry: pointer to dentry of the base directory
1681 * @mnt: pointer to vfs mount of the base directory
1682 * @name: pointer to file name
1683 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001684 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001685 */
1686int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1687 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001688 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001689{
Al Viroe0a01242011-06-27 17:00:37 -04001690 struct nameidata nd;
1691 int err;
1692 nd.root.dentry = dentry;
1693 nd.root.mnt = mnt;
1694 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001695 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001696 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1697 if (!err)
1698 *path = nd.path;
1699 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001700}
1701
Christoph Hellwigeead1912007-10-16 23:25:38 -07001702static struct dentry *__lookup_hash(struct qstr *name,
1703 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001705 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001706 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 int err;
1708
Al Viro4ad5abb2011-06-20 19:57:03 -04001709 err = inode_permission(inode, MAY_EXEC);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001710 if (err)
1711 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
1713 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001714 * Don't bother with __d_lookup: callers are for creat as
1715 * well as unlink, so a lot of the time it would cost
1716 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001717 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001718 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001719
Josef Bacik44396f42011-05-31 11:58:49 -04001720 if (dentry && d_need_lookup(dentry)) {
1721 /*
1722 * __lookup_hash is called with the parent dir's i_mutex already
1723 * held, so we are good to go here.
1724 */
1725 dentry = d_inode_lookup(base, dentry, nd);
1726 if (IS_ERR(dentry))
1727 return dentry;
1728 }
1729
Al Virod2d9e9f2011-06-20 10:55:26 -04001730 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1731 int status = d_revalidate(dentry, nd);
1732 if (unlikely(status <= 0)) {
1733 /*
1734 * The dentry failed validation.
1735 * If d_revalidate returned 0 attempt to invalidate
1736 * the dentry otherwise d_revalidate is asking us
1737 * to return a fail status.
1738 */
1739 if (status < 0) {
1740 dput(dentry);
1741 return ERR_PTR(status);
1742 } else if (!d_invalidate(dentry)) {
1743 dput(dentry);
1744 dentry = NULL;
1745 }
1746 }
1747 }
Al Viro6e6b1bd2009-08-13 23:38:37 +04001748
Nick Pigginbaa03892010-08-18 04:37:31 +10001749 if (!dentry)
1750 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return dentry;
1753}
1754
James Morris057f6c02007-04-26 00:12:05 -07001755/*
1756 * Restricted form of lookup. Doesn't follow links, single-component only,
1757 * needs parent already locked. Doesn't follow mounts.
1758 * SMP-safe.
1759 */
Adrian Bunka244e162006-03-31 02:32:11 -08001760static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761{
Jan Blunck4ac91372008-02-14 19:34:32 -08001762 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763}
1764
Christoph Hellwigeead1912007-10-16 23:25:38 -07001765/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001766 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001767 * @name: pathname component to lookup
1768 * @base: base directory to lookup from
1769 * @len: maximum length @len should be interpreted to
1770 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001771 * Note that this routine is purely a helper for filesystem usage and should
1772 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001773 * nameidata argument is passed to the filesystem methods and a filesystem
1774 * using this helper needs to be prepared for that.
1775 */
James Morris057f6c02007-04-26 00:12:05 -07001776struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1777{
James Morris057f6c02007-04-26 00:12:05 -07001778 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001779 unsigned long hash;
1780 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001781
David Woodhouse2f9092e2009-04-20 23:18:37 +01001782 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1783
Al Viro6a96ba52011-03-07 23:49:20 -05001784 this.name = name;
1785 this.len = len;
1786 if (!len)
1787 return ERR_PTR(-EACCES);
1788
1789 hash = init_name_hash();
1790 while (len--) {
1791 c = *(const unsigned char *)name++;
1792 if (c == '/' || c == '\0')
1793 return ERR_PTR(-EACCES);
1794 hash = partial_name_hash(c, hash);
1795 }
1796 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001797 /*
1798 * See if the low-level filesystem might want
1799 * to use its own hash..
1800 */
1801 if (base->d_flags & DCACHE_OP_HASH) {
1802 int err = base->d_op->d_hash(base, base->d_inode, &this);
1803 if (err < 0)
1804 return ERR_PTR(err);
1805 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001806
Christoph Hellwig49705b72005-11-08 21:35:06 -08001807 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001808}
1809
Al Viro2d8f3032008-07-22 09:59:21 -04001810int user_path_at(int dfd, const char __user *name, unsigned flags,
1811 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812{
Al Viro2d8f3032008-07-22 09:59:21 -04001813 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001814 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001817
1818 BUG_ON(flags & LOOKUP_PARENT);
1819
1820 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001822 if (!err)
1823 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 }
1825 return err;
1826}
1827
Al Viro2ad94ae2008-07-21 09:32:51 -04001828static int user_path_parent(int dfd, const char __user *path,
1829 struct nameidata *nd, char **name)
1830{
1831 char *s = getname(path);
1832 int error;
1833
1834 if (IS_ERR(s))
1835 return PTR_ERR(s);
1836
1837 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1838 if (error)
1839 putname(s);
1840 else
1841 *name = s;
1842
1843 return error;
1844}
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846/*
1847 * It's inline, so penalty for filesystems that don't use sticky bit is
1848 * minimal.
1849 */
1850static inline int check_sticky(struct inode *dir, struct inode *inode)
1851{
David Howellsda9592e2008-11-14 10:39:05 +11001852 uid_t fsuid = current_fsuid();
1853
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 if (!(dir->i_mode & S_ISVTX))
1855 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001856 if (current_user_ns() != inode_userns(inode))
1857 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001858 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001860 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001862
1863other_userns:
1864 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865}
1866
1867/*
1868 * Check whether we can remove a link victim from directory dir, check
1869 * whether the type of victim is right.
1870 * 1. We can't do it if dir is read-only (done in permission())
1871 * 2. We should have write and exec permissions on dir
1872 * 3. We can't remove anything from append-only dir
1873 * 4. We can't do anything with immutable dir (done in permission())
1874 * 5. If the sticky bit on dir is set we should either
1875 * a. be owner of dir, or
1876 * b. be owner of victim, or
1877 * c. have CAP_FOWNER capability
1878 * 6. If the victim is append-only or immutable we can't do antyhing with
1879 * links pointing to it.
1880 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1881 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1882 * 9. We can't remove a root or mountpoint.
1883 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1884 * nfs_async_unlink().
1885 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001886static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887{
1888 int error;
1889
1890 if (!victim->d_inode)
1891 return -ENOENT;
1892
1893 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001894 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Al Virof419a2e2008-07-22 00:07:17 -04001896 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 if (error)
1898 return error;
1899 if (IS_APPEND(dir))
1900 return -EPERM;
1901 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001902 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 return -EPERM;
1904 if (isdir) {
1905 if (!S_ISDIR(victim->d_inode->i_mode))
1906 return -ENOTDIR;
1907 if (IS_ROOT(victim))
1908 return -EBUSY;
1909 } else if (S_ISDIR(victim->d_inode->i_mode))
1910 return -EISDIR;
1911 if (IS_DEADDIR(dir))
1912 return -ENOENT;
1913 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1914 return -EBUSY;
1915 return 0;
1916}
1917
1918/* Check whether we can create an object with dentry child in directory
1919 * dir.
1920 * 1. We can't do it if child already exists (open has special treatment for
1921 * this case, but since we are inlined it's OK)
1922 * 2. We can't do it if dir is read-only (done in permission())
1923 * 3. We should have write and exec permissions on dir
1924 * 4. We can't do it if dir is immutable (done in permission())
1925 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001926static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927{
1928 if (child->d_inode)
1929 return -EEXIST;
1930 if (IS_DEADDIR(dir))
1931 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001932 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933}
1934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935/*
1936 * p1 and p2 should be directories on the same fs.
1937 */
1938struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1939{
1940 struct dentry *p;
1941
1942 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 return NULL;
1945 }
1946
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001947 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001949 p = d_ancestor(p2, p1);
1950 if (p) {
1951 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1952 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1953 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 }
1955
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001956 p = d_ancestor(p1, p2);
1957 if (p) {
1958 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1959 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1960 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 }
1962
Ingo Molnarf2eace22006-07-03 00:25:05 -07001963 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1964 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return NULL;
1966}
1967
1968void unlock_rename(struct dentry *p1, struct dentry *p2)
1969{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001970 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001972 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001973 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 }
1975}
1976
1977int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1978 struct nameidata *nd)
1979{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001980 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
1982 if (error)
1983 return error;
1984
Al Viroacfa4382008-12-04 10:06:33 -05001985 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 return -EACCES; /* shouldn't it be ENOSYS? */
1987 mode &= S_IALLUGO;
1988 mode |= S_IFREG;
1989 error = security_inode_create(dir, dentry, mode);
1990 if (error)
1991 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001993 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001994 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 return error;
1996}
1997
Al Viro73d049a2011-03-11 12:08:24 -05001998static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002000 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 struct inode *inode = dentry->d_inode;
2002 int error;
2003
Al Virobcda7652011-03-13 16:42:14 -04002004 /* O_PATH? */
2005 if (!acc_mode)
2006 return 0;
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 if (!inode)
2009 return -ENOENT;
2010
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002011 switch (inode->i_mode & S_IFMT) {
2012 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002014 case S_IFDIR:
2015 if (acc_mode & MAY_WRITE)
2016 return -EISDIR;
2017 break;
2018 case S_IFBLK:
2019 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002020 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002022 /*FALLTHRU*/
2023 case S_IFIFO:
2024 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002026 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002027 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002028
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002029 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002030 if (error)
2031 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002032
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 /*
2034 * An append-only file must be opened in append mode for writing.
2035 */
2036 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002037 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002038 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002040 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042
2043 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002044 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002045 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 /*
2048 * Ensure there are no outstanding leases on the file.
2049 */
Al Virob65a9cf2009-12-16 06:27:40 -05002050 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002051}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Jeff Laytone1181ee2010-12-07 16:19:50 -05002053static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002054{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002055 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002056 struct inode *inode = path->dentry->d_inode;
2057 int error = get_write_access(inode);
2058 if (error)
2059 return error;
2060 /*
2061 * Refuse to truncate files with mandatory locks held on them.
2062 */
2063 error = locks_verify_locked(inode);
2064 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002065 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002066 if (!error) {
2067 error = do_truncate(path->dentry, 0,
2068 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002069 filp);
Al Viro7715b522009-12-16 03:54:00 -05002070 }
2071 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002072 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073}
2074
Dave Hansend57999e2008-02-15 14:37:27 -08002075static inline int open_to_namei_flags(int flag)
2076{
Al Viro8a5e9292011-06-25 19:15:54 -04002077 if ((flag & O_ACCMODE) == 3)
2078 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002079 return flag;
2080}
2081
Nick Piggin31e6b012011-01-07 17:49:52 +11002082/*
Al Virofe2d35f2011-03-05 22:58:25 -05002083 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002084 */
Al Virofb1cc552009-12-24 01:58:28 -05002085static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002086 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002087{
Al Viroa1e28032009-12-24 02:12:06 -05002088 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002089 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002090 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002091 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002092 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002093 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002094 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002095 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002096
Al Viroc3e380b2011-02-23 13:39:45 -05002097 nd->flags &= ~LOOKUP_PARENT;
2098 nd->flags |= op->intent;
2099
Al Viro1f36f772009-12-26 10:56:19 -05002100 switch (nd->last_type) {
2101 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002102 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002103 error = handle_dots(nd, nd->last_type);
2104 if (error)
2105 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002106 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002107 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002108 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002109 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002110 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002111 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002112 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002113 error = -EISDIR;
2114 goto exit;
2115 }
2116 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002117 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002118 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002119 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002120 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002121 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002122 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002123 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002124
Al Viroca344a892011-03-09 00:36:45 -05002125 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002126 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002127 if (nd->last.name[nd->last.len])
2128 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002129 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2130 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002131 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002132 error = walk_component(nd, path, &nd->last, LAST_NORM,
2133 !symlink_ok);
2134 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002135 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002136 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002137 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002138 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002139 error = complete_walk(nd);
2140 if (error)
2141 return ERR_PTR(-ECHILD);
Al Virofe2d35f2011-03-05 22:58:25 -05002142
2143 error = -ENOTDIR;
2144 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002145 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002146 goto exit;
2147 }
2148 audit_inode(pathname, nd->path.dentry);
2149 goto ok;
2150 }
2151
2152 /* create side of things */
Al Viro9f1fafe2011-03-25 11:00:12 -04002153 error = complete_walk(nd);
2154 if (error)
2155 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002156
2157 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002158 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002159 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002160 if (nd->last.name[nd->last.len])
2161 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002162
Al Viroa1e28032009-12-24 02:12:06 -05002163 mutex_lock(&dir->d_inode->i_mutex);
2164
Al Viro6c0d46c2011-03-09 00:59:59 -05002165 dentry = lookup_hash(nd);
2166 error = PTR_ERR(dentry);
2167 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002168 mutex_unlock(&dir->d_inode->i_mutex);
2169 goto exit;
2170 }
2171
Al Viro6c0d46c2011-03-09 00:59:59 -05002172 path->dentry = dentry;
2173 path->mnt = nd->path.mnt;
2174
Al Virofb1cc552009-12-24 01:58:28 -05002175 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002176 if (!dentry->d_inode) {
2177 int mode = op->mode;
2178 if (!IS_POSIXACL(dir->d_inode))
2179 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002180 /*
2181 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002182 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002183 * the time when the file is created and when
2184 * a permanent write count is taken through
2185 * the 'struct file' in nameidata_to_filp().
2186 */
2187 error = mnt_want_write(nd->path.mnt);
2188 if (error)
2189 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002190 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002191 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002192 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002193 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002194 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002195 error = security_path_mknod(&nd->path, dentry, mode, 0);
2196 if (error)
2197 goto exit_mutex_unlock;
2198 error = vfs_create(dir->d_inode, dentry, mode, nd);
2199 if (error)
2200 goto exit_mutex_unlock;
2201 mutex_unlock(&dir->d_inode->i_mutex);
2202 dput(nd->path.dentry);
2203 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002204 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002205 }
2206
2207 /*
2208 * It already exists.
2209 */
2210 mutex_unlock(&dir->d_inode->i_mutex);
2211 audit_inode(pathname, path->dentry);
2212
2213 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002214 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002215 goto exit_dput;
2216
David Howells9875cf82011-01-14 18:45:21 +00002217 error = follow_managed(path, nd->flags);
2218 if (error < 0)
2219 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002220
2221 error = -ENOENT;
2222 if (!path->dentry->d_inode)
2223 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002224
2225 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002226 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002227
2228 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002229 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002230 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002231 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002232 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002233ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002234 if (!S_ISREG(nd->inode->i_mode))
2235 will_truncate = 0;
2236
Al Viro0f9d1a12011-03-09 00:13:14 -05002237 if (will_truncate) {
2238 error = mnt_want_write(nd->path.mnt);
2239 if (error)
2240 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002241 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002242 }
Al Viroca344a892011-03-09 00:36:45 -05002243common:
Al Virobcda7652011-03-13 16:42:14 -04002244 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002245 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002246 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002247 filp = nameidata_to_filp(nd);
2248 if (!IS_ERR(filp)) {
2249 error = ima_file_check(filp, op->acc_mode);
2250 if (error) {
2251 fput(filp);
2252 filp = ERR_PTR(error);
2253 }
2254 }
2255 if (!IS_ERR(filp)) {
2256 if (will_truncate) {
2257 error = handle_truncate(filp);
2258 if (error) {
2259 fput(filp);
2260 filp = ERR_PTR(error);
2261 }
2262 }
2263 }
Al Viroca344a892011-03-09 00:36:45 -05002264out:
2265 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002266 mnt_drop_write(nd->path.mnt);
2267 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002268 return filp;
2269
2270exit_mutex_unlock:
2271 mutex_unlock(&dir->d_inode->i_mutex);
2272exit_dput:
2273 path_put_conditional(path, nd);
2274exit:
Al Viroca344a892011-03-09 00:36:45 -05002275 filp = ERR_PTR(error);
2276 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002277}
2278
Al Viro13aab422011-02-23 17:54:08 -05002279static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002280 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281{
Al Virofe2d35f2011-03-05 22:58:25 -05002282 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002283 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002284 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002285 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002286
2287 filp = get_empty_filp();
2288 if (!filp)
2289 return ERR_PTR(-ENFILE);
2290
Al Viro47c805d2011-02-23 17:44:09 -05002291 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002292 nd->intent.open.file = filp;
2293 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2294 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002295
Al Viro73d049a2011-03-11 12:08:24 -05002296 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002297 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002298 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002299
Al Virofe2d35f2011-03-05 22:58:25 -05002300 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002301 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002302 if (unlikely(error))
2303 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
Al Viro73d049a2011-03-11 12:08:24 -05002305 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002306 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002307 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002308 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002309 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002310 path_put_conditional(&path, nd);
2311 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002312 filp = ERR_PTR(-ELOOP);
2313 break;
2314 }
Al Viro73d049a2011-03-11 12:08:24 -05002315 nd->flags |= LOOKUP_PARENT;
2316 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002317 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002318 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002319 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002320 else
Al Viro73d049a2011-03-11 12:08:24 -05002321 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002322 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002323 }
Al Viro10fa8e62009-12-26 07:09:49 -05002324out:
Al Viro73d049a2011-03-11 12:08:24 -05002325 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2326 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002327 if (base)
2328 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002329 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002330 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Nick Piggin31e6b012011-01-07 17:49:52 +11002332out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002333 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002334 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335}
2336
Al Viro13aab422011-02-23 17:54:08 -05002337struct file *do_filp_open(int dfd, const char *pathname,
2338 const struct open_flags *op, int flags)
2339{
Al Viro73d049a2011-03-11 12:08:24 -05002340 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002341 struct file *filp;
2342
Al Viro73d049a2011-03-11 12:08:24 -05002343 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002344 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002345 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002346 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002347 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002348 return filp;
2349}
2350
Al Viro73d049a2011-03-11 12:08:24 -05002351struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2352 const char *name, const struct open_flags *op, int flags)
2353{
2354 struct nameidata nd;
2355 struct file *file;
2356
2357 nd.root.mnt = mnt;
2358 nd.root.dentry = dentry;
2359
2360 flags |= LOOKUP_ROOT;
2361
Al Virobcda7652011-03-13 16:42:14 -04002362 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002363 return ERR_PTR(-ELOOP);
2364
2365 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2366 if (unlikely(file == ERR_PTR(-ECHILD)))
2367 file = path_openat(-1, name, &nd, op, flags);
2368 if (unlikely(file == ERR_PTR(-ESTALE)))
2369 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2370 return file;
2371}
2372
Al Viroed75e952011-06-27 16:53:43 -04002373struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002375 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002376 struct nameidata nd;
2377 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2378 if (error)
2379 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002381 /*
2382 * Yucky last component or no last component at all?
2383 * (foo/., foo/.., /////)
2384 */
Al Viroed75e952011-06-27 16:53:43 -04002385 if (nd.last_type != LAST_NORM)
2386 goto out;
2387 nd.flags &= ~LOOKUP_PARENT;
2388 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2389 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002390
2391 /*
2392 * Do the final lookup.
2393 */
Al Viroed75e952011-06-27 16:53:43 -04002394 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2395 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 if (IS_ERR(dentry))
2397 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002398
Al Viroe9baf6e2008-05-15 04:49:12 -04002399 if (dentry->d_inode)
2400 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002401 /*
2402 * Special case - lookup gave negative, but... we had foo/bar/
2403 * From the vfs_mknod() POV we just have a negative dentry -
2404 * all is fine. Let's be bastards - you had / on the end, you've
2405 * been asking for (non-existent) directory. -ENOENT for you.
2406 */
Al Viroed75e952011-06-27 16:53:43 -04002407 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002408 dput(dentry);
2409 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002410 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002411 }
Al Viroed75e952011-06-27 16:53:43 -04002412 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002414eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002416 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417fail:
Al Viroed75e952011-06-27 16:53:43 -04002418 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2419out:
2420 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 return dentry;
2422}
Al Virodae6ad82011-06-26 11:50:15 -04002423EXPORT_SYMBOL(kern_path_create);
2424
2425struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2426{
2427 char *tmp = getname(pathname);
2428 struct dentry *res;
2429 if (IS_ERR(tmp))
2430 return ERR_CAST(tmp);
2431 res = kern_path_create(dfd, tmp, path, is_dir);
2432 putname(tmp);
2433 return res;
2434}
2435EXPORT_SYMBOL(user_path_create);
2436
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2438{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002439 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440
2441 if (error)
2442 return error;
2443
Serge E. Hallyne795b712011-03-23 16:43:25 -07002444 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2445 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 return -EPERM;
2447
Al Viroacfa4382008-12-04 10:06:33 -05002448 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 return -EPERM;
2450
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002451 error = devcgroup_inode_mknod(mode, dev);
2452 if (error)
2453 return error;
2454
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 error = security_inode_mknod(dir, dentry, mode, dev);
2456 if (error)
2457 return error;
2458
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002460 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002461 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 return error;
2463}
2464
Dave Hansen463c3192008-02-15 14:37:57 -08002465static int may_mknod(mode_t mode)
2466{
2467 switch (mode & S_IFMT) {
2468 case S_IFREG:
2469 case S_IFCHR:
2470 case S_IFBLK:
2471 case S_IFIFO:
2472 case S_IFSOCK:
2473 case 0: /* zero mode translates to S_IFREG */
2474 return 0;
2475 case S_IFDIR:
2476 return -EPERM;
2477 default:
2478 return -EINVAL;
2479 }
2480}
2481
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002482SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2483 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484{
Al Viro2ad94ae2008-07-21 09:32:51 -04002485 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002486 struct path path;
2487 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
2489 if (S_ISDIR(mode))
2490 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Al Virodae6ad82011-06-26 11:50:15 -04002492 dentry = user_path_create(dfd, filename, &path, 0);
2493 if (IS_ERR(dentry))
2494 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002495
Al Virodae6ad82011-06-26 11:50:15 -04002496 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002497 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002498 error = may_mknod(mode);
2499 if (error)
2500 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002501 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002502 if (error)
2503 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002504 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002505 if (error)
2506 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002507 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002509 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 break;
2511 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002512 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 new_decode_dev(dev));
2514 break;
2515 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002516 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002519out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002520 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002521out_dput:
2522 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002523 mutex_unlock(&path.dentry->d_inode->i_mutex);
2524 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
2526 return error;
2527}
2528
Heiko Carstens3480b252009-01-14 14:14:16 +01002529SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002530{
2531 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2532}
2533
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2535{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002536 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
2538 if (error)
2539 return error;
2540
Al Viroacfa4382008-12-04 10:06:33 -05002541 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 return -EPERM;
2543
2544 mode &= (S_IRWXUGO|S_ISVTX);
2545 error = security_inode_mkdir(dir, dentry, mode);
2546 if (error)
2547 return error;
2548
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002550 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002551 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return error;
2553}
2554
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002555SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556{
Dave Hansen6902d922006-09-30 23:29:01 -07002557 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002558 struct path path;
2559 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Al Virodae6ad82011-06-26 11:50:15 -04002561 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002562 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002563 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002564
Al Virodae6ad82011-06-26 11:50:15 -04002565 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002566 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002567 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002568 if (error)
2569 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002570 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002571 if (error)
2572 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002573 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002574out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002575 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002576out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002577 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002578 mutex_unlock(&path.dentry->d_inode->i_mutex);
2579 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 return error;
2581}
2582
Heiko Carstens3cdad422009-01-14 14:14:22 +01002583SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002584{
2585 return sys_mkdirat(AT_FDCWD, pathname, mode);
2586}
2587
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588/*
Sage Weila71905f2011-05-24 13:06:08 -07002589 * The dentry_unhash() helper will try to drop the dentry early: we
2590 * should have a usage count of 2 if we're the only user of this
2591 * dentry, and if that is true (possibly after pruning the dcache),
2592 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 *
2594 * A low-level filesystem can, if it choses, legally
2595 * do a
2596 *
2597 * if (!d_unhashed(dentry))
2598 * return -EBUSY;
2599 *
2600 * if it cannot handle the case of removing a directory
2601 * that is still in use by something else..
2602 */
2603void dentry_unhash(struct dentry *dentry)
2604{
Vasily Averindc168422006-12-06 20:37:07 -08002605 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002607 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 __d_drop(dentry);
2609 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610}
2611
2612int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2613{
2614 int error = may_delete(dir, dentry, 1);
2615
2616 if (error)
2617 return error;
2618
Al Viroacfa4382008-12-04 10:06:33 -05002619 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 return -EPERM;
2621
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002622 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
Sage Weil912dbc12011-05-24 13:06:11 -07002624 error = -EBUSY;
2625 if (d_mountpoint(dentry))
2626 goto out;
2627
2628 error = security_inode_rmdir(dir, dentry);
2629 if (error)
2630 goto out;
2631
Sage Weil3cebde22011-05-29 21:20:59 -07002632 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002633 error = dir->i_op->rmdir(dir, dentry);
2634 if (error)
2635 goto out;
2636
2637 dentry->d_inode->i_flags |= S_DEAD;
2638 dont_mount(dentry);
2639
2640out:
2641 mutex_unlock(&dentry->d_inode->i_mutex);
2642 if (!error)
2643 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 return error;
2645}
2646
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002647static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648{
2649 int error = 0;
2650 char * name;
2651 struct dentry *dentry;
2652 struct nameidata nd;
2653
Al Viro2ad94ae2008-07-21 09:32:51 -04002654 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002656 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
2658 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002659 case LAST_DOTDOT:
2660 error = -ENOTEMPTY;
2661 goto exit1;
2662 case LAST_DOT:
2663 error = -EINVAL;
2664 goto exit1;
2665 case LAST_ROOT:
2666 error = -EBUSY;
2667 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002669
2670 nd.flags &= ~LOOKUP_PARENT;
2671
Jan Blunck4ac91372008-02-14 19:34:32 -08002672 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002673 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002675 if (IS_ERR(dentry))
2676 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002677 if (!dentry->d_inode) {
2678 error = -ENOENT;
2679 goto exit3;
2680 }
Dave Hansen06227532008-02-15 14:37:34 -08002681 error = mnt_want_write(nd.path.mnt);
2682 if (error)
2683 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002684 error = security_path_rmdir(&nd.path, dentry);
2685 if (error)
2686 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002687 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002688exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002689 mnt_drop_write(nd.path.mnt);
2690exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002691 dput(dentry);
2692exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002693 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002695 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 putname(name);
2697 return error;
2698}
2699
Heiko Carstens3cdad422009-01-14 14:14:22 +01002700SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002701{
2702 return do_rmdir(AT_FDCWD, pathname);
2703}
2704
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705int vfs_unlink(struct inode *dir, struct dentry *dentry)
2706{
2707 int error = may_delete(dir, dentry, 0);
2708
2709 if (error)
2710 return error;
2711
Al Viroacfa4382008-12-04 10:06:33 -05002712 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 return -EPERM;
2714
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002715 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 if (d_mountpoint(dentry))
2717 error = -EBUSY;
2718 else {
2719 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002720 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002722 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002723 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002726 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
2728 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2729 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002730 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 }
Robert Love0eeca282005-07-12 17:06:03 -04002733
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 return error;
2735}
2736
2737/*
2738 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002739 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 * writeout happening, and we don't want to prevent access to the directory
2741 * while waiting on the I/O.
2742 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002743static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744{
Al Viro2ad94ae2008-07-21 09:32:51 -04002745 int error;
2746 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 struct dentry *dentry;
2748 struct nameidata nd;
2749 struct inode *inode = NULL;
2750
Al Viro2ad94ae2008-07-21 09:32:51 -04002751 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002753 return error;
2754
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 error = -EISDIR;
2756 if (nd.last_type != LAST_NORM)
2757 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002758
2759 nd.flags &= ~LOOKUP_PARENT;
2760
Jan Blunck4ac91372008-02-14 19:34:32 -08002761 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002762 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 error = PTR_ERR(dentry);
2764 if (!IS_ERR(dentry)) {
2765 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002766 if (nd.last.name[nd.last.len])
2767 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002769 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002770 goto slashes;
2771 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002772 error = mnt_want_write(nd.path.mnt);
2773 if (error)
2774 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002775 error = security_path_unlink(&nd.path, dentry);
2776 if (error)
2777 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002778 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002779exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002780 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 exit2:
2782 dput(dentry);
2783 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002784 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 if (inode)
2786 iput(inode); /* truncate the inode here */
2787exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002788 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 putname(name);
2790 return error;
2791
2792slashes:
2793 error = !dentry->d_inode ? -ENOENT :
2794 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2795 goto exit2;
2796}
2797
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002798SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002799{
2800 if ((flag & ~AT_REMOVEDIR) != 0)
2801 return -EINVAL;
2802
2803 if (flag & AT_REMOVEDIR)
2804 return do_rmdir(dfd, pathname);
2805
2806 return do_unlinkat(dfd, pathname);
2807}
2808
Heiko Carstens3480b252009-01-14 14:14:16 +01002809SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002810{
2811 return do_unlinkat(AT_FDCWD, pathname);
2812}
2813
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002814int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002816 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
2818 if (error)
2819 return error;
2820
Al Viroacfa4382008-12-04 10:06:33 -05002821 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 return -EPERM;
2823
2824 error = security_inode_symlink(dir, dentry, oldname);
2825 if (error)
2826 return error;
2827
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002829 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002830 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 return error;
2832}
2833
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002834SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2835 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836{
Al Viro2ad94ae2008-07-21 09:32:51 -04002837 int error;
2838 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07002839 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002840 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
2842 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002843 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002845
Al Virodae6ad82011-06-26 11:50:15 -04002846 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07002847 error = PTR_ERR(dentry);
2848 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002849 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07002850
Al Virodae6ad82011-06-26 11:50:15 -04002851 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002852 if (error)
2853 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002854 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002855 if (error)
2856 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002857 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002858out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002859 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002860out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002861 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002862 mutex_unlock(&path.dentry->d_inode->i_mutex);
2863 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07002864out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 putname(from);
2866 return error;
2867}
2868
Heiko Carstens3480b252009-01-14 14:14:16 +01002869SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002870{
2871 return sys_symlinkat(oldname, AT_FDCWD, newname);
2872}
2873
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2875{
2876 struct inode *inode = old_dentry->d_inode;
2877 int error;
2878
2879 if (!inode)
2880 return -ENOENT;
2881
Miklos Szeredia95164d2008-07-30 15:08:48 +02002882 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 if (error)
2884 return error;
2885
2886 if (dir->i_sb != inode->i_sb)
2887 return -EXDEV;
2888
2889 /*
2890 * A link to an append-only or immutable file cannot be created.
2891 */
2892 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2893 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002894 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002896 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 return -EPERM;
2898
2899 error = security_inode_link(old_dentry, dir, new_dentry);
2900 if (error)
2901 return error;
2902
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002903 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302904 /* Make sure we don't allow creating hardlink to an unlinked file */
2905 if (inode->i_nlink == 0)
2906 error = -ENOENT;
2907 else
2908 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002909 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002910 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002911 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 return error;
2913}
2914
2915/*
2916 * Hardlinks are often used in delicate situations. We avoid
2917 * security-related surprises by not following symlinks on the
2918 * newname. --KAB
2919 *
2920 * We don't follow them on the oldname either to be compatible
2921 * with linux 2.0, and to avoid hard-linking to directories
2922 * and other special files. --ADM
2923 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002924SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2925 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926{
2927 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002928 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302929 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302932 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002933 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302934 /*
2935 * To use null names we require CAP_DAC_READ_SEARCH
2936 * This ensures that not everyone will be able to create
2937 * handlink using the passed filedescriptor.
2938 */
2939 if (flags & AT_EMPTY_PATH) {
2940 if (!capable(CAP_DAC_READ_SEARCH))
2941 return -ENOENT;
2942 how = LOOKUP_EMPTY;
2943 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002944
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302945 if (flags & AT_SYMLINK_FOLLOW)
2946 how |= LOOKUP_FOLLOW;
2947
2948 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002950 return error;
2951
Al Virodae6ad82011-06-26 11:50:15 -04002952 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002954 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002955 goto out;
2956
2957 error = -EXDEV;
2958 if (old_path.mnt != new_path.mnt)
2959 goto out_dput;
2960 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002961 if (error)
2962 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002963 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002964 if (error)
2965 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002966 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002967out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002968 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002969out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002970 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002971 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2972 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973out:
Al Viro2d8f3032008-07-22 09:59:21 -04002974 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
2976 return error;
2977}
2978
Heiko Carstens3480b252009-01-14 14:14:16 +01002979SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002980{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002981 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002982}
2983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984/*
2985 * The worst of all namespace operations - renaming directory. "Perverted"
2986 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2987 * Problems:
2988 * a) we can get into loop creation. Check is done in is_subdir().
2989 * b) race potential - two innocent renames can create a loop together.
2990 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002991 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 * story.
2993 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002994 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 * whether the target exists). Solution: try to be smart with locking
2996 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002997 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 * move will be locked. Thus we can rank directories by the tree
2999 * (ancestors first) and rank all non-directories after them.
3000 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003001 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 * HOWEVER, it relies on the assumption that any object with ->lookup()
3003 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3004 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003005 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003006 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003008 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 * locking].
3010 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003011static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3012 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013{
3014 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003015 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
3017 /*
3018 * If we are going to change the parent - check write permissions,
3019 * we'll need to flip '..'.
3020 */
3021 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003022 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 if (error)
3024 return error;
3025 }
3026
3027 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3028 if (error)
3029 return error;
3030
Al Virod83c49f2010-04-30 17:17:09 -04003031 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003032 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003033
3034 error = -EBUSY;
3035 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3036 goto out;
3037
Sage Weil3cebde22011-05-29 21:20:59 -07003038 if (target)
3039 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003040 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3041 if (error)
3042 goto out;
3043
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003045 target->i_flags |= S_DEAD;
3046 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 }
Sage Weil9055cba2011-05-24 13:06:12 -07003048out:
3049 if (target)
3050 mutex_unlock(&target->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003051 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003052 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3053 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 return error;
3055}
3056
Adrian Bunk75c96f82005-05-05 16:16:09 -07003057static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3058 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059{
Sage Weil51892bb2011-05-24 13:06:13 -07003060 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 int error;
3062
3063 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3064 if (error)
3065 return error;
3066
3067 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003069 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003070
3071 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003073 goto out;
3074
3075 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3076 if (error)
3077 goto out;
3078
3079 if (target)
3080 dont_mount(new_dentry);
3081 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3082 d_move(old_dentry, new_dentry);
3083out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003085 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 dput(new_dentry);
3087 return error;
3088}
3089
3090int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3091 struct inode *new_dir, struct dentry *new_dentry)
3092{
3093 int error;
3094 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003095 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
3097 if (old_dentry->d_inode == new_dentry->d_inode)
3098 return 0;
3099
3100 error = may_delete(old_dir, old_dentry, is_dir);
3101 if (error)
3102 return error;
3103
3104 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003105 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 else
3107 error = may_delete(new_dir, new_dentry, is_dir);
3108 if (error)
3109 return error;
3110
Al Viroacfa4382008-12-04 10:06:33 -05003111 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 return -EPERM;
3113
Robert Love0eeca282005-07-12 17:06:03 -04003114 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3115
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 if (is_dir)
3117 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3118 else
3119 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003120 if (!error)
3121 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003122 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003123 fsnotify_oldname_free(old_name);
3124
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 return error;
3126}
3127
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003128SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3129 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130{
Al Viro2ad94ae2008-07-21 09:32:51 -04003131 struct dentry *old_dir, *new_dir;
3132 struct dentry *old_dentry, *new_dentry;
3133 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003135 char *from;
3136 char *to;
3137 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138
Al Viro2ad94ae2008-07-21 09:32:51 -04003139 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 if (error)
3141 goto exit;
3142
Al Viro2ad94ae2008-07-21 09:32:51 -04003143 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 if (error)
3145 goto exit1;
3146
3147 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003148 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 goto exit2;
3150
Jan Blunck4ac91372008-02-14 19:34:32 -08003151 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 error = -EBUSY;
3153 if (oldnd.last_type != LAST_NORM)
3154 goto exit2;
3155
Jan Blunck4ac91372008-02-14 19:34:32 -08003156 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 if (newnd.last_type != LAST_NORM)
3158 goto exit2;
3159
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003160 oldnd.flags &= ~LOOKUP_PARENT;
3161 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003162 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003163
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 trap = lock_rename(new_dir, old_dir);
3165
Christoph Hellwig49705b72005-11-08 21:35:06 -08003166 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 error = PTR_ERR(old_dentry);
3168 if (IS_ERR(old_dentry))
3169 goto exit3;
3170 /* source must exist */
3171 error = -ENOENT;
3172 if (!old_dentry->d_inode)
3173 goto exit4;
3174 /* unless the source is a directory trailing slashes give -ENOTDIR */
3175 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3176 error = -ENOTDIR;
3177 if (oldnd.last.name[oldnd.last.len])
3178 goto exit4;
3179 if (newnd.last.name[newnd.last.len])
3180 goto exit4;
3181 }
3182 /* source should not be ancestor of target */
3183 error = -EINVAL;
3184 if (old_dentry == trap)
3185 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003186 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 error = PTR_ERR(new_dentry);
3188 if (IS_ERR(new_dentry))
3189 goto exit4;
3190 /* target should not be an ancestor of source */
3191 error = -ENOTEMPTY;
3192 if (new_dentry == trap)
3193 goto exit5;
3194
Dave Hansen9079b1e2008-02-15 14:37:49 -08003195 error = mnt_want_write(oldnd.path.mnt);
3196 if (error)
3197 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003198 error = security_path_rename(&oldnd.path, old_dentry,
3199 &newnd.path, new_dentry);
3200 if (error)
3201 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202 error = vfs_rename(old_dir->d_inode, old_dentry,
3203 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003204exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003205 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206exit5:
3207 dput(new_dentry);
3208exit4:
3209 dput(old_dentry);
3210exit3:
3211 unlock_rename(new_dir, old_dir);
3212exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003213 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003214 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003216 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003218exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 return error;
3220}
3221
Heiko Carstensa26eab22009-01-14 14:14:17 +01003222SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003223{
3224 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3225}
3226
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3228{
3229 int len;
3230
3231 len = PTR_ERR(link);
3232 if (IS_ERR(link))
3233 goto out;
3234
3235 len = strlen(link);
3236 if (len > (unsigned) buflen)
3237 len = buflen;
3238 if (copy_to_user(buffer, link, len))
3239 len = -EFAULT;
3240out:
3241 return len;
3242}
3243
3244/*
3245 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3246 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3247 * using) it for any given inode is up to filesystem.
3248 */
3249int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3250{
3251 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003252 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003253 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003254
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003256 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003257 if (IS_ERR(cookie))
3258 return PTR_ERR(cookie);
3259
3260 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3261 if (dentry->d_inode->i_op->put_link)
3262 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3263 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264}
3265
3266int vfs_follow_link(struct nameidata *nd, const char *link)
3267{
3268 return __vfs_follow_link(nd, link);
3269}
3270
3271/* get the link contents into pagecache */
3272static char *page_getlink(struct dentry * dentry, struct page **ppage)
3273{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003274 char *kaddr;
3275 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003277 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003279 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003281 kaddr = kmap(page);
3282 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3283 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284}
3285
3286int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3287{
3288 struct page *page = NULL;
3289 char *s = page_getlink(dentry, &page);
3290 int res = vfs_readlink(dentry,buffer,buflen,s);
3291 if (page) {
3292 kunmap(page);
3293 page_cache_release(page);
3294 }
3295 return res;
3296}
3297
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003298void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003300 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003302 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303}
3304
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003305void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003307 struct page *page = cookie;
3308
3309 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 kunmap(page);
3311 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 }
3313}
3314
Nick Piggin54566b22009-01-04 12:00:53 -08003315/*
3316 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3317 */
3318int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319{
3320 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003321 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003322 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003323 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003325 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3326 if (nofs)
3327 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328
NeilBrown7e53cac2006-03-25 03:07:57 -08003329retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003330 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003331 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003333 goto fail;
3334
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 kaddr = kmap_atomic(page, KM_USER0);
3336 memcpy(kaddr, symname, len-1);
3337 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003338
3339 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3340 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 if (err < 0)
3342 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003343 if (err < len-1)
3344 goto retry;
3345
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 mark_inode_dirty(inode);
3347 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348fail:
3349 return err;
3350}
3351
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003352int page_symlink(struct inode *inode, const char *symname, int len)
3353{
3354 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003355 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003356}
3357
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003358const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 .readlink = generic_readlink,
3360 .follow_link = page_follow_link_light,
3361 .put_link = page_put_link,
3362};
3363
Al Viro2d8f3032008-07-22 09:59:21 -04003364EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003365EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366EXPORT_SYMBOL(follow_down);
3367EXPORT_SYMBOL(follow_up);
3368EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3369EXPORT_SYMBOL(getname);
3370EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371EXPORT_SYMBOL(lookup_one_len);
3372EXPORT_SYMBOL(page_follow_link_light);
3373EXPORT_SYMBOL(page_put_link);
3374EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003375EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376EXPORT_SYMBOL(page_symlink);
3377EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003378EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003379EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003380EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381EXPORT_SYMBOL(unlock_rename);
3382EXPORT_SYMBOL(vfs_create);
3383EXPORT_SYMBOL(vfs_follow_link);
3384EXPORT_SYMBOL(vfs_link);
3385EXPORT_SYMBOL(vfs_mkdir);
3386EXPORT_SYMBOL(vfs_mknod);
3387EXPORT_SYMBOL(generic_permission);
3388EXPORT_SYMBOL(vfs_readlink);
3389EXPORT_SYMBOL(vfs_rename);
3390EXPORT_SYMBOL(vfs_rmdir);
3391EXPORT_SYMBOL(vfs_symlink);
3392EXPORT_SYMBOL(vfs_unlink);
3393EXPORT_SYMBOL(dentry_unhash);
3394EXPORT_SYMBOL(generic_readlink);