blob: 46ea9cc16647c063ba0f485cf838d566e6b3dd42 [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"
Al Viroc7105362011-11-24 18:22:03 -050039#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
46 *
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
53 *
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
57 *
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
60 *
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
67 */
68
69/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030075 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 *
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
84 */
85
86/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
88 *
89 * [10-Sep-98 Alan Modra] Another symlink change.
90 */
91
92/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
99 *
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
105 */
106/*
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 * any extra contention...
110 */
111
112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800119static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100141static char *getname_flags(const char __user *filename, int flags, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
Eric Paris4043cde2012-01-03 14:23:08 -0500143 char *result = __getname();
144 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Eric Paris4043cde2012-01-03 14:23:08 -0500146 if (!result)
147 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Eric Paris4043cde2012-01-03 14:23:08 -0500149 retval = do_getname(filename, result);
150 if (retval < 0) {
151 if (retval == -ENOENT && empty)
152 *empty = 1;
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(result);
155 return ERR_PTR(retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 }
157 }
158 audit_getname(result);
159 return result;
160}
161
Al Virof52e0c12011-03-14 18:56:51 -0400162char *getname(const char __user * filename)
163{
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100164 return getname_flags(filename, 0, 0);
Al Virof52e0c12011-03-14 18:56:51 -0400165}
166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167#ifdef CONFIG_AUDITSYSCALL
168void putname(const char *name)
169{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400170 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 audit_putname(name);
172 else
173 __putname(name);
174}
175EXPORT_SYMBOL(putname);
176#endif
177
Linus Torvaldse77819e2011-07-22 19:30:19 -0700178static int check_acl(struct inode *inode, int mask)
179{
Linus Torvalds84635d62011-07-25 22:47:03 -0700180#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700181 struct posix_acl *acl;
182
Linus Torvaldse77819e2011-07-22 19:30:19 -0700183 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
185 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700186 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
189 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700191 }
192
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
194
195 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700199 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700202 */
203 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
206 if (IS_ERR(acl))
207 return PTR_ERR(acl);
208 } else {
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
210 return -EAGAIN;
211 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700212 }
213
214 if (acl) {
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
217 return error;
218 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700219#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700220
221 return -EAGAIN;
222}
223
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700224/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530225 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700226 */
Al Viro7e401452011-06-20 19:12:17 -0400227static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700228{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700229 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700230
Serge E. Hallyne795b712011-03-23 16:43:25 -0700231 if (current_user_ns() != inode_userns(inode))
232 goto other_perms;
233
Linus Torvalds14067ff2011-07-25 19:55:52 -0700234 if (likely(current_fsuid() == inode->i_uid))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700235 mode >>= 6;
236 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700237 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400238 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100239 if (error != -EAGAIN)
240 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700241 }
242
243 if (in_group_p(inode->i_gid))
244 mode >>= 3;
245 }
246
Serge E. Hallyne795b712011-03-23 16:43:25 -0700247other_perms:
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700248 /*
249 * If the DACs are ok we don't need any capability check.
250 */
Al Viro9c2c7032011-06-20 19:06:22 -0400251 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700252 return 0;
253 return -EACCES;
254}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100257 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 *
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100264 * are used for other things.
265 *
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
Al Viro2830ba72011-06-20 19:16:29 -0400270int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700272 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530275 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 */
Al Viro7e401452011-06-20 19:12:17 -0400277 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700278 if (ret != -EACCES)
279 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Al Virod594e7e2011-06-20 19:55:42 -0400281 if (S_ISDIR(inode->i_mode)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
284 return 0;
285 if (!(mask & MAY_WRITE))
286 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
287 return 0;
288 return -EACCES;
289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 /*
291 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 */
Al Virod594e7e2011-06-20 19:55:42 -0400295 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700296 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return 0;
298
299 /*
300 * Searching includes executable on directories, else just read.
301 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600302 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400303 if (mask == MAY_READ)
Serge E. Hallyne795b712011-03-23 16:43:25 -0700304 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 return 0;
306
307 return -EACCES;
308}
309
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700310/*
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
315 */
316static inline int do_inode_permission(struct inode *inode, int mask)
317{
318 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
319 if (likely(inode->i_op->permission))
320 return inode->i_op->permission(inode, mask);
321
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode->i_lock);
324 inode->i_opflags |= IOP_FASTPERM;
325 spin_unlock(&inode->i_lock);
326 }
327 return generic_permission(inode, mask);
328}
329
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200330/**
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200334 *
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530339 *
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200341 */
Al Virof419a2e2008-07-22 00:07:17 -0400342int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Al Viroe6305c42008-07-15 21:03:57 -0400344 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700346 if (unlikely(mask & MAY_WRITE)) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700347 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 /*
350 * Nobody gets write access to a read-only fs.
351 */
352 if (IS_RDONLY(inode) &&
353 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
354 return -EROFS;
355
356 /*
357 * Nobody gets write access to an immutable file.
358 */
359 if (IS_IMMUTABLE(inode))
360 return -EACCES;
361 }
362
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700363 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 if (retval)
365 return retval;
366
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700367 retval = devcgroup_inode_permission(inode, mask);
368 if (retval)
369 return retval;
370
Eric Parisd09ca732010-07-23 11:43:57 -0400371 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
Al Virof4d6ff82011-06-19 13:14:21 -0400374/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800375 * path_get - get a reference to a path
376 * @path: path to get the reference to
377 *
378 * Given a path increment the reference count to the dentry and the vfsmount.
379 */
380void path_get(struct path *path)
381{
382 mntget(path->mnt);
383 dget(path->dentry);
384}
385EXPORT_SYMBOL(path_get);
386
387/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800388 * path_put - put a reference to a path
389 * @path: path to put the reference to
390 *
391 * Given a path decrement the reference count to the dentry and the vfsmount.
392 */
393void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Jan Blunck1d957f92008-02-14 19:34:35 -0800395 dput(path->dentry);
396 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
Jan Blunck1d957f92008-02-14 19:34:35 -0800398EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Al Viro19660af2011-03-25 10:32:48 -0400400/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100401 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100409 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100410
411/**
Al Viro19660af2011-03-25 10:32:48 -0400412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800415 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100416 *
Al Viro19660af2011-03-25 10:32:48 -0400417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100420 */
Al Viro19660af2011-03-25 10:32:48 -0400421static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100422{
423 struct fs_struct *fs = current->fs;
424 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500425 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100426
427 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500428 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
429 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100430 spin_lock(&fs->lock);
431 if (nd->root.mnt != fs->root.mnt ||
432 nd->root.dentry != fs->root.dentry)
433 goto err_root;
434 }
435 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400436 if (!dentry) {
437 if (!__d_rcu_to_refcount(parent, nd->seq))
438 goto err_parent;
439 BUG_ON(nd->inode != parent->d_inode);
440 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400441 if (dentry->d_parent != parent)
442 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400443 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
444 if (!__d_rcu_to_refcount(dentry, nd->seq))
445 goto err_child;
446 /*
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
451 */
452 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
453 BUG_ON(!parent->d_count);
454 parent->d_count++;
455 spin_unlock(&dentry->d_lock);
456 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100457 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500458 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100459 path_get(&nd->root);
460 spin_unlock(&fs->lock);
461 }
462 mntget(nd->path.mnt);
463
464 rcu_read_unlock();
465 br_read_unlock(vfsmount_lock);
466 nd->flags &= ~LOOKUP_RCU;
467 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400468
469err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100470 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400471err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100472 spin_unlock(&parent->d_lock);
473err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500474 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100475 spin_unlock(&fs->lock);
476 return -ECHILD;
477}
478
Nick Piggin31e6b012011-01-07 17:49:52 +1100479/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
482 */
483void release_open_intent(struct nameidata *nd)
484{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800485 struct file *file = nd->intent.open.file;
486
487 if (file && !IS_ERR(file)) {
488 if (file->f_path.dentry == NULL)
489 put_filp(file);
490 else
491 fput(file);
492 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700493}
494
Al Virof60aef72011-02-15 01:35:28 -0500495static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100496{
Al Virof60aef72011-02-15 01:35:28 -0500497 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100498}
499
Al Viro9f1fafe2011-03-25 11:00:12 -0400500/**
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500503 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500509 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400510static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500511{
Al Viro16c2cd72011-02-22 15:50:10 -0500512 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500513 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500514
Al Viro9f1fafe2011-03-25 11:00:12 -0400515 if (nd->flags & LOOKUP_RCU) {
516 nd->flags &= ~LOOKUP_RCU;
517 if (!(nd->flags & LOOKUP_ROOT))
518 nd->root.mnt = NULL;
519 spin_lock(&dentry->d_lock);
520 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
521 spin_unlock(&dentry->d_lock);
522 rcu_read_unlock();
523 br_read_unlock(vfsmount_lock);
524 return -ECHILD;
525 }
526 BUG_ON(nd->inode != dentry->d_inode);
527 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
529 rcu_read_unlock();
530 br_read_unlock(vfsmount_lock);
531 }
532
Al Viro16c2cd72011-02-22 15:50:10 -0500533 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500534 return 0;
535
Al Viro16c2cd72011-02-22 15:50:10 -0500536 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
537 return 0;
538
539 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
540 return 0;
541
542 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100543 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500544 if (status > 0)
545 return 0;
546
Al Viro16c2cd72011-02-22 15:50:10 -0500547 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500548 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500549
Al Viro9f1fafe2011-03-25 11:00:12 -0400550 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500551 return status;
552}
553
Al Viro2a737872009-04-07 11:49:53 -0400554static __always_inline void set_root(struct nameidata *nd)
555{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200556 if (!nd->root.mnt)
557 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400558}
559
Al Viro6de88d72009-08-09 01:41:57 +0400560static int link_path_walk(const char *, struct nameidata *);
561
Nick Piggin31e6b012011-01-07 17:49:52 +1100562static __always_inline void set_root_rcu(struct nameidata *nd)
563{
564 if (!nd->root.mnt) {
565 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100566 unsigned seq;
567
568 do {
569 seq = read_seqcount_begin(&fs->seq);
570 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700571 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100572 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100573 }
574}
575
Arjan van de Venf1662352006-01-14 13:21:31 -0800576static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
Nick Piggin31e6b012011-01-07 17:49:52 +1100578 int ret;
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (IS_ERR(link))
581 goto fail;
582
583 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400584 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800585 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400586 nd->path = nd->root;
587 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500588 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100590 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100591
Nick Piggin31e6b012011-01-07 17:49:52 +1100592 ret = link_path_walk(link, nd);
593 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800595 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return PTR_ERR(link);
597}
598
Jan Blunck1d957f92008-02-14 19:34:35 -0800599static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700600{
601 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800602 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700603 mntput(path->mnt);
604}
605
Nick Piggin7b9337a2011-01-14 08:42:43 +0000606static inline void path_to_nameidata(const struct path *path,
607 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700608{
Nick Piggin31e6b012011-01-07 17:49:52 +1100609 if (!(nd->flags & LOOKUP_RCU)) {
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800613 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100614 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800615 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700616}
617
Al Viro574197e2011-03-14 22:20:34 -0400618static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
619{
620 struct inode *inode = link->dentry->d_inode;
621 if (!IS_ERR(cookie) && inode->i_op->put_link)
622 inode->i_op->put_link(link->dentry, nd, cookie);
623 path_put(link);
624}
625
Al Virodef4af32009-12-26 08:37:05 -0500626static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400627follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800628{
629 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000630 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800631
Al Viro844a3912011-02-15 00:38:26 -0500632 BUG_ON(nd->flags & LOOKUP_RCU);
633
Al Viro0e794582011-03-16 02:45:02 -0400634 if (link->mnt == nd->path.mnt)
635 mntget(link->mnt);
636
Al Viro574197e2011-03-14 22:20:34 -0400637 if (unlikely(current->total_link_count >= 40)) {
638 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400639 path_put(&nd->path);
640 return -ELOOP;
641 }
642 cond_resched();
643 current->total_link_count++;
644
Nick Piggin7b9337a2011-01-14 08:42:43 +0000645 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800646 nd_set_link(nd, NULL);
647
Al Viro36f3b4f2011-02-22 21:24:38 -0500648 error = security_inode_follow_link(link->dentry, nd);
649 if (error) {
650 *p = ERR_PTR(error); /* no ->put_link(), please */
651 path_put(&nd->path);
652 return error;
653 }
654
Al Viro86acdca12009-12-22 23:45:11 -0500655 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500656 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
657 error = PTR_ERR(*p);
658 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800659 char *s = nd_get_link(nd);
660 error = 0;
661 if (s)
662 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400663 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500664 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400665 nd->inode = nd->path.dentry->d_inode;
666 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400667 /* stepped on a _really_ weird one */
668 path_put(&nd->path);
669 error = -ELOOP;
670 }
671 }
Ian Kent051d3812006-03-27 01:14:53 -0800672 }
Ian Kent051d3812006-03-27 01:14:53 -0800673 return error;
674}
675
Nick Piggin31e6b012011-01-07 17:49:52 +1100676static int follow_up_rcu(struct path *path)
677{
Al Viro0714a532011-11-24 22:19:58 -0500678 struct mount *mnt = real_mount(path->mnt);
679 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100680 struct dentry *mountpoint;
681
Al Viro0714a532011-11-24 22:19:58 -0500682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100684 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500685 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100686 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500687 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100688 return 1;
689}
690
Al Virobab77eb2009-04-18 03:26:48 -0400691int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Al Viro0714a532011-11-24 22:19:58 -0500693 struct mount *mnt = real_mount(path->mnt);
694 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000696
697 br_read_lock(vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000700 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return 0;
702 }
Al Viro0714a532011-11-24 22:19:58 -0500703 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500704 mountpoint = dget(mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000705 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400706 dput(path->dentry);
707 path->dentry = mountpoint;
708 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500709 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 return 1;
711}
712
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100713/*
David Howells9875cf82011-01-14 18:45:21 +0000714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
716 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 */
David Howells9875cf82011-01-14 18:45:21 +0000718static int follow_automount(struct path *path, unsigned flags,
719 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100720{
David Howells9875cf82011-01-14 18:45:21 +0000721 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000722 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100723
David Howells9875cf82011-01-14 18:45:21 +0000724 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
725 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700726
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
729 * the name.
730 *
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000737 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200738 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700739 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200740 path->dentry->d_inode)
741 return -EISDIR;
742
David Howells9875cf82011-01-14 18:45:21 +0000743 current->total_link_count++;
744 if (current->total_link_count >= 40)
745 return -ELOOP;
746
747 mnt = path->dentry->d_op->d_automount(path);
748 if (IS_ERR(mnt)) {
749 /*
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
753 *
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
757 */
Al Viro49084c32011-06-25 21:59:52 -0400758 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000759 return -EREMOTE;
760 return PTR_ERR(mnt);
761 }
David Howellsea5b7782011-01-14 19:10:03 +0000762
David Howells9875cf82011-01-14 18:45:21 +0000763 if (!mnt) /* mount collision */
764 return 0;
765
Al Viro8aef1882011-06-16 15:10:06 +0100766 if (!*need_mntput) {
767 /* lock_mount() may release path->mnt on error */
768 mntget(path->mnt);
769 *need_mntput = true;
770 }
Al Viro19a167a2011-01-17 01:35:23 -0500771 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000772
David Howellsea5b7782011-01-14 19:10:03 +0000773 switch (err) {
774 case -EBUSY:
775 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500776 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000777 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100778 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000779 path->mnt = mnt;
780 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000781 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500782 default:
783 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000784 }
Al Viro19a167a2011-01-17 01:35:23 -0500785
David Howells9875cf82011-01-14 18:45:21 +0000786}
787
788/*
789 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000790 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000791 * - Flagged as mountpoint
792 * - Flagged as automount point
793 *
794 * This may only be called in refwalk mode.
795 *
796 * Serialization is taken care of in namespace.c
797 */
798static int follow_managed(struct path *path, unsigned flags)
799{
Al Viro8aef1882011-06-16 15:10:06 +0100800 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000801 unsigned managed;
802 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100803 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000804
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed = ACCESS_ONCE(path->dentry->d_flags),
809 managed &= DCACHE_MANAGED_DENTRY,
810 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000811 /* Allow the filesystem to manage the transit without i_mutex
812 * being held. */
813 if (managed & DCACHE_MANAGE_TRANSIT) {
814 BUG_ON(!path->dentry->d_op);
815 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400816 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000817 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100818 break;
David Howellscc53ce52011-01-14 18:45:26 +0000819 }
820
David Howells9875cf82011-01-14 18:45:21 +0000821 /* Transit to a mounted filesystem. */
822 if (managed & DCACHE_MOUNTED) {
823 struct vfsmount *mounted = lookup_mnt(path);
824 if (mounted) {
825 dput(path->dentry);
826 if (need_mntput)
827 mntput(path->mnt);
828 path->mnt = mounted;
829 path->dentry = dget(mounted->mnt_root);
830 need_mntput = true;
831 continue;
832 }
833
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
837 * vfsmount_lock */
838 }
839
840 /* Handle an automount point */
841 if (managed & DCACHE_NEED_AUTOMOUNT) {
842 ret = follow_automount(path, flags, &need_mntput);
843 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100844 break;
David Howells9875cf82011-01-14 18:45:21 +0000845 continue;
846 }
847
848 /* We didn't change the current path point */
849 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
Al Viro8aef1882011-06-16 15:10:06 +0100851
852 if (need_mntput && path->mnt == mnt)
853 mntput(path->mnt);
854 if (ret == -EISDIR)
855 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +0000856 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857}
858
David Howellscc53ce52011-01-14 18:45:26 +0000859int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 struct vfsmount *mounted;
862
Al Viro1c755af2009-04-18 14:06:57 -0400863 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400865 dput(path->dentry);
866 mntput(path->mnt);
867 path->mnt = mounted;
868 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 return 1;
870 }
871 return 0;
872}
873
Ian Kent62a73752011-03-25 01:51:02 +0800874static inline bool managed_dentry_might_block(struct dentry *dentry)
875{
876 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
877 dentry->d_op->d_manage(dentry, true) < 0);
878}
879
David Howells9875cf82011-01-14 18:45:21 +0000880/*
Al Viro287548e2011-05-27 06:50:06 -0400881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000883 */
884static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400885 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000886{
Ian Kent62a73752011-03-25 01:51:02 +0800887 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -0500888 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800889 /*
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
892 */
Al Viro287548e2011-05-27 06:50:06 -0400893 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000894 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800895
896 if (!d_mountpoint(path->dentry))
897 break;
898
David Howells9875cf82011-01-14 18:45:21 +0000899 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
900 if (!mounted)
901 break;
Al Viroc7105362011-11-24 18:22:03 -0500902 path->mnt = &mounted->mnt;
903 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +0000904 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +0000905 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700906 /*
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
910 */
911 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000912 }
David Howells9875cf82011-01-14 18:45:21 +0000913 return true;
914}
915
Al Virodea39372011-05-27 06:53:39 -0400916static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400917{
Al Virodea39372011-05-27 06:53:39 -0400918 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -0500919 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400920 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400921 if (!mounted)
922 break;
Al Viroc7105362011-11-24 18:22:03 -0500923 nd->path.mnt = &mounted->mnt;
924 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -0400925 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400926 }
927}
928
Nick Piggin31e6b012011-01-07 17:49:52 +1100929static int follow_dotdot_rcu(struct nameidata *nd)
930{
Nick Piggin31e6b012011-01-07 17:49:52 +1100931 set_root_rcu(nd);
932
David Howells9875cf82011-01-14 18:45:21 +0000933 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100934 if (nd->path.dentry == nd->root.dentry &&
935 nd->path.mnt == nd->root.mnt) {
936 break;
937 }
938 if (nd->path.dentry != nd->path.mnt->mnt_root) {
939 struct dentry *old = nd->path.dentry;
940 struct dentry *parent = old->d_parent;
941 unsigned seq;
942
943 seq = read_seqcount_begin(&parent->d_seq);
944 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500945 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100946 nd->path.dentry = parent;
947 nd->seq = seq;
948 break;
949 }
950 if (!follow_up_rcu(&nd->path))
951 break;
952 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100953 }
Al Virodea39372011-05-27 06:53:39 -0400954 follow_mount_rcu(nd);
955 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100956 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500957
958failed:
959 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500960 if (!(nd->flags & LOOKUP_ROOT))
961 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500962 rcu_read_unlock();
963 br_read_unlock(vfsmount_lock);
964 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100965}
966
David Howells9875cf82011-01-14 18:45:21 +0000967/*
David Howellscc53ce52011-01-14 18:45:26 +0000968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +0000971 */
Al Viro7cc90cc2011-03-18 09:04:20 -0400972int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +0000973{
974 unsigned managed;
975 int ret;
976
977 while (managed = ACCESS_ONCE(path->dentry->d_flags),
978 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
979 /* Allow the filesystem to manage the transit without i_mutex
980 * being held.
981 *
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
985 * superstructure.
986 *
987 * The filesystem may sleep at this point.
988 */
989 if (managed & DCACHE_MANAGE_TRANSIT) {
990 BUG_ON(!path->dentry->d_op);
991 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000992 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -0400993 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000994 if (ret < 0)
995 return ret == -EISDIR ? 0 : ret;
996 }
997
998 /* Transit to a mounted filesystem. */
999 if (managed & DCACHE_MOUNTED) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1001 if (!mounted)
1002 break;
1003 dput(path->dentry);
1004 mntput(path->mnt);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1007 continue;
1008 }
1009
1010 /* Don't handle automount points here */
1011 break;
1012 }
1013 return 0;
1014}
1015
1016/*
David Howells9875cf82011-01-14 18:45:21 +00001017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1018 */
1019static void follow_mount(struct path *path)
1020{
1021 while (d_mountpoint(path->dentry)) {
1022 struct vfsmount *mounted = lookup_mnt(path);
1023 if (!mounted)
1024 break;
1025 dput(path->dentry);
1026 mntput(path->mnt);
1027 path->mnt = mounted;
1028 path->dentry = dget(mounted->mnt_root);
1029 }
1030}
1031
Nick Piggin31e6b012011-01-07 17:49:52 +11001032static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Al Viro2a737872009-04-07 11:49:53 -04001034 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001037 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Al Viro2a737872009-04-07 11:49:53 -04001039 if (nd->path.dentry == nd->root.dentry &&
1040 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 break;
1042 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001043 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001044 /* rare case of legitimate dget_parent()... */
1045 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 dput(old);
1047 break;
1048 }
Al Viro3088dd72010-01-30 15:47:29 -05001049 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
Al Viro79ed0222009-04-18 13:59:41 -04001052 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001053 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001057 * Allocate a dentry with name and parent, and perform a parent
1058 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1059 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1060 * have verified that no child exists while under i_mutex.
1061 */
1062static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1063 struct qstr *name, struct nameidata *nd)
1064{
1065 struct inode *inode = parent->d_inode;
1066 struct dentry *dentry;
1067 struct dentry *old;
1068
1069 /* Don't create child dentry for a dead directory. */
1070 if (unlikely(IS_DEADDIR(inode)))
1071 return ERR_PTR(-ENOENT);
1072
1073 dentry = d_alloc(parent, name);
1074 if (unlikely(!dentry))
1075 return ERR_PTR(-ENOMEM);
1076
1077 old = inode->i_op->lookup(inode, dentry, nd);
1078 if (unlikely(old)) {
1079 dput(dentry);
1080 dentry = old;
1081 }
1082 return dentry;
1083}
1084
1085/*
Josef Bacik44396f42011-05-31 11:58:49 -04001086 * We already have a dentry, but require a lookup to be performed on the parent
1087 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1088 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1089 * child exists while under i_mutex.
1090 */
1091static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1092 struct nameidata *nd)
1093{
1094 struct inode *inode = parent->d_inode;
1095 struct dentry *old;
1096
1097 /* Don't create child dentry for a dead directory. */
Miklos Szeredie188dc02012-02-03 14:25:18 +01001098 if (unlikely(IS_DEADDIR(inode))) {
1099 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001100 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001101 }
Josef Bacik44396f42011-05-31 11:58:49 -04001102
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1105 dput(dentry);
1106 dentry = old;
1107 }
1108 return dentry;
1109}
1110
1111/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 * It's more convoluted than I'd like it to be, but... it's still fairly
1113 * small and for now I'd prefer to have fast path as straight as possible.
1114 * It _is_ time-critical.
1115 */
1116static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001117 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118{
Jan Blunck4ac91372008-02-14 19:34:32 -08001119 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001120 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001121 int need_reval = 1;
1122 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001123 int err;
1124
Al Viro3cac2602009-08-13 18:27:43 +04001125 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001126 * Rename seqlock is not required here because in the off chance
1127 * of a false negative due to a concurrent rename, we're going to
1128 * do the non-racy lookup, below.
1129 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001130 if (nd->flags & LOOKUP_RCU) {
1131 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001132 *inode = nd->inode;
1133 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001134 if (!dentry)
1135 goto unlazy;
1136
Nick Piggin31e6b012011-01-07 17:49:52 +11001137 /* Memory barrier in read_seqcount_begin of child is enough */
1138 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1139 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001140 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001141
Al Viro24643082011-02-15 01:26:22 -05001142 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001143 status = d_revalidate(dentry, nd);
1144 if (unlikely(status <= 0)) {
1145 if (status != -ECHILD)
1146 need_reval = 0;
1147 goto unlazy;
1148 }
Al Viro24643082011-02-15 01:26:22 -05001149 }
Josef Bacik44396f42011-05-31 11:58:49 -04001150 if (unlikely(d_need_lookup(dentry)))
1151 goto unlazy;
Nick Piggin31e6b012011-01-07 17:49:52 +11001152 path->mnt = mnt;
1153 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001154 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1155 goto unlazy;
1156 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1157 goto unlazy;
1158 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001159unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001160 if (unlazy_walk(nd, dentry))
1161 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001162 } else {
1163 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001164 }
Al Viro5a18fff2011-03-11 04:44:53 -05001165
Josef Bacik44396f42011-05-31 11:58:49 -04001166 if (dentry && unlikely(d_need_lookup(dentry))) {
1167 dput(dentry);
1168 dentry = NULL;
1169 }
Al Viro5a18fff2011-03-11 04:44:53 -05001170retry:
1171 if (unlikely(!dentry)) {
1172 struct inode *dir = parent->d_inode;
1173 BUG_ON(nd->inode != dir);
1174
1175 mutex_lock(&dir->i_mutex);
1176 dentry = d_lookup(parent, name);
1177 if (likely(!dentry)) {
1178 dentry = d_alloc_and_lookup(parent, name, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1182 }
1183 /* known good */
1184 need_reval = 0;
1185 status = 1;
Josef Bacik44396f42011-05-31 11:58:49 -04001186 } else if (unlikely(d_need_lookup(dentry))) {
1187 dentry = d_inode_lookup(parent, dentry, nd);
1188 if (IS_ERR(dentry)) {
1189 mutex_unlock(&dir->i_mutex);
1190 return PTR_ERR(dentry);
1191 }
1192 /* known good */
1193 need_reval = 0;
1194 status = 1;
Al Viro5a18fff2011-03-11 04:44:53 -05001195 }
1196 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001197 }
Al Viro5a18fff2011-03-11 04:44:53 -05001198 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1199 status = d_revalidate(dentry, nd);
1200 if (unlikely(status <= 0)) {
1201 if (status < 0) {
1202 dput(dentry);
1203 return status;
1204 }
1205 if (!d_invalidate(dentry)) {
1206 dput(dentry);
1207 dentry = NULL;
1208 need_reval = 1;
1209 goto retry;
1210 }
1211 }
1212
David Howells9875cf82011-01-14 18:45:21 +00001213 path->mnt = mnt;
1214 path->dentry = dentry;
1215 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001216 if (unlikely(err < 0)) {
1217 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001218 return err;
Ian Kent89312212011-01-18 12:06:10 +08001219 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001220 if (err)
1221 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001222 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
1225
Al Viro52094c82011-02-21 21:34:47 -05001226static inline int may_lookup(struct nameidata *nd)
1227{
1228 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001229 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001230 if (err != -ECHILD)
1231 return err;
Al Viro19660af2011-03-25 10:32:48 -04001232 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001233 return -ECHILD;
1234 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001235 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001236}
1237
Al Viro9856fa12011-03-04 14:22:06 -05001238static inline int handle_dots(struct nameidata *nd, int type)
1239{
1240 if (type == LAST_DOTDOT) {
1241 if (nd->flags & LOOKUP_RCU) {
1242 if (follow_dotdot_rcu(nd))
1243 return -ECHILD;
1244 } else
1245 follow_dotdot(nd);
1246 }
1247 return 0;
1248}
1249
Al Viro951361f2011-03-04 14:44:37 -05001250static void terminate_walk(struct nameidata *nd)
1251{
1252 if (!(nd->flags & LOOKUP_RCU)) {
1253 path_put(&nd->path);
1254 } else {
1255 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001256 if (!(nd->flags & LOOKUP_ROOT))
1257 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001258 rcu_read_unlock();
1259 br_read_unlock(vfsmount_lock);
1260 }
1261}
1262
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001263/*
1264 * Do we need to follow links? We _really_ want to be able
1265 * to do this check without having to look at inode->i_op,
1266 * so we keep a cache of "no, this doesn't need follow_link"
1267 * for the common case.
1268 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001269static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001270{
1271 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1272 if (likely(inode->i_op->follow_link))
1273 return follow;
1274
1275 /* This gets set once for the inode lifetime */
1276 spin_lock(&inode->i_lock);
1277 inode->i_opflags |= IOP_NOFOLLOW;
1278 spin_unlock(&inode->i_lock);
1279 }
1280 return 0;
1281}
1282
Al Viroce57dfc2011-03-13 19:58:58 -04001283static inline int walk_component(struct nameidata *nd, struct path *path,
1284 struct qstr *name, int type, int follow)
1285{
1286 struct inode *inode;
1287 int err;
1288 /*
1289 * "." and ".." are special - ".." especially so because it has
1290 * to be able to know about the current root directory and
1291 * parent relationships.
1292 */
1293 if (unlikely(type != LAST_NORM))
1294 return handle_dots(nd, type);
1295 err = do_lookup(nd, name, path, &inode);
1296 if (unlikely(err)) {
1297 terminate_walk(nd);
1298 return err;
1299 }
1300 if (!inode) {
1301 path_to_nameidata(path, nd);
1302 terminate_walk(nd);
1303 return -ENOENT;
1304 }
Linus Torvalds7813b942011-08-07 09:53:20 -07001305 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001306 if (nd->flags & LOOKUP_RCU) {
1307 if (unlikely(unlazy_walk(nd, path->dentry))) {
1308 terminate_walk(nd);
1309 return -ECHILD;
1310 }
1311 }
Al Viroce57dfc2011-03-13 19:58:58 -04001312 BUG_ON(inode != path->dentry->d_inode);
1313 return 1;
1314 }
1315 path_to_nameidata(path, nd);
1316 nd->inode = inode;
1317 return 0;
1318}
1319
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320/*
Al Virob3563792011-03-14 21:54:55 -04001321 * This limits recursive symlink follows to 8, while
1322 * limiting consecutive symlinks to 40.
1323 *
1324 * Without that kind of total limit, nasty chains of consecutive
1325 * symlinks can cause almost arbitrarily long lookups.
1326 */
1327static inline int nested_symlink(struct path *path, struct nameidata *nd)
1328{
1329 int res;
1330
Al Virob3563792011-03-14 21:54:55 -04001331 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1332 path_put_conditional(path, nd);
1333 path_put(&nd->path);
1334 return -ELOOP;
1335 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001336 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001337
1338 nd->depth++;
1339 current->link_count++;
1340
1341 do {
1342 struct path link = *path;
1343 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001344
1345 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001346 if (!res)
1347 res = walk_component(nd, path, &nd->last,
1348 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001349 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001350 } while (res > 0);
1351
1352 current->link_count--;
1353 nd->depth--;
1354 return res;
1355}
1356
1357/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001358 * We really don't want to look at inode->i_op->lookup
1359 * when we don't have to. So we keep a cache bit in
1360 * the inode ->i_opflags field that says "yes, we can
1361 * do lookup on this inode".
1362 */
1363static inline int can_lookup(struct inode *inode)
1364{
1365 if (likely(inode->i_opflags & IOP_LOOKUP))
1366 return 1;
1367 if (likely(!inode->i_op->lookup))
1368 return 0;
1369
1370 /* We do this once for the lifetime of the inode */
1371 spin_lock(&inode->i_lock);
1372 inode->i_opflags |= IOP_LOOKUP;
1373 spin_unlock(&inode->i_lock);
1374 return 1;
1375}
1376
Linus Torvalds0145acc2012-03-02 14:32:59 -08001377unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1378{
1379 unsigned long hash = init_name_hash();
1380 while (len--)
1381 hash = partial_name_hash(*name++, hash);
1382 return end_name_hash(hash);
1383}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001384EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001385
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001386/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001387 * We know there's a real path component here of at least
1388 * one character.
1389 */
1390static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1391{
1392 unsigned long hash = init_name_hash();
1393 unsigned long len = 0, c;
1394
1395 c = (unsigned char)*name;
1396 do {
1397 len++;
1398 hash = partial_name_hash(c, hash);
1399 c = (unsigned char)name[len];
1400 } while (c && c != '/');
1401 *hashp = end_name_hash(hash);
1402 return len;
1403}
1404
1405/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001407 * This is the basic name resolution function, turning a pathname into
1408 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001410 * Returns 0 and nd will have valid dentry and mnt on success.
1411 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 */
Al Viro6de88d72009-08-09 01:41:57 +04001413static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414{
1415 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
1418 while (*name=='/')
1419 name++;
1420 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001421 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 /* At this point we know we have a real path component. */
1424 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001426 long len;
Al Virofe479a52011-02-22 15:10:03 -05001427 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Al Viro52094c82011-02-21 21:34:47 -05001429 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (err)
1431 break;
1432
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001433 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001435 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
Al Virofe479a52011-02-22 15:10:03 -05001437 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001438 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001439 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001440 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001441 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001442 nd->flags |= LOOKUP_JUMPED;
1443 }
Al Virofe479a52011-02-22 15:10:03 -05001444 break;
1445 case 1:
1446 type = LAST_DOT;
1447 }
Al Viro5a202bc2011-03-08 14:17:44 -05001448 if (likely(type == LAST_NORM)) {
1449 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001450 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001451 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1452 err = parent->d_op->d_hash(parent, nd->inode,
1453 &this);
1454 if (err < 0)
1455 break;
1456 }
1457 }
Al Virofe479a52011-02-22 15:10:03 -05001458
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001459 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001461 /*
1462 * If it wasn't NUL, we know it was '/'. Skip that
1463 * slash, and continue until no more slashes.
1464 */
1465 do {
1466 len++;
1467 } while (unlikely(name[len] == '/'));
1468 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001469 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001470 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Al Viroce57dfc2011-03-13 19:58:58 -04001472 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1473 if (err < 0)
1474 return err;
Al Virofe479a52011-02-22 15:10:03 -05001475
Al Viroce57dfc2011-03-13 19:58:58 -04001476 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001477 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001479 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001480 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001481 if (can_lookup(nd->inode))
1482 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001484 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 /* here ends the main loop */
1486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487last_component:
Al Virob3563792011-03-14 21:54:55 -04001488 nd->last = this;
1489 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001490 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 }
Al Viro951361f2011-03-04 14:44:37 -05001492 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return err;
1494}
1495
Al Viro70e9b352011-03-05 21:12:22 -05001496static int path_init(int dfd, const char *name, unsigned int flags,
1497 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001499 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001500 int fput_needed;
1501 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001504 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001506 if (flags & LOOKUP_ROOT) {
1507 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001508 if (*name) {
1509 if (!inode->i_op->lookup)
1510 return -ENOTDIR;
1511 retval = inode_permission(inode, MAY_EXEC);
1512 if (retval)
1513 return retval;
1514 }
Al Viro5b6ca022011-03-09 23:04:47 -05001515 nd->path = nd->root;
1516 nd->inode = inode;
1517 if (flags & LOOKUP_RCU) {
1518 br_read_lock(vfsmount_lock);
1519 rcu_read_lock();
1520 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1521 } else {
1522 path_get(&nd->path);
1523 }
1524 return 0;
1525 }
1526
Al Viro2a737872009-04-07 11:49:53 -04001527 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001530 if (flags & LOOKUP_RCU) {
1531 br_read_lock(vfsmount_lock);
1532 rcu_read_lock();
1533 set_root_rcu(nd);
1534 } else {
1535 set_root(nd);
1536 path_get(&nd->root);
1537 }
Al Viro2a737872009-04-07 11:49:53 -04001538 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001539 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001540 if (flags & LOOKUP_RCU) {
1541 struct fs_struct *fs = current->fs;
1542 unsigned seq;
1543
1544 br_read_lock(vfsmount_lock);
1545 rcu_read_lock();
1546
1547 do {
1548 seq = read_seqcount_begin(&fs->seq);
1549 nd->path = fs->pwd;
1550 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1551 } while (read_seqcount_retry(&fs->seq, seq));
1552 } else {
1553 get_fs_pwd(current->fs, &nd->path);
1554 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001555 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001556 struct dentry *dentry;
1557
Al Viro1abf0c72011-03-13 03:51:11 -04001558 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001559 retval = -EBADF;
1560 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001561 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001562
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001563 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001564
Al Virof52e0c12011-03-14 18:56:51 -04001565 if (*name) {
1566 retval = -ENOTDIR;
1567 if (!S_ISDIR(dentry->d_inode->i_mode))
1568 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001569
Al Viro4ad5abb2011-06-20 19:57:03 -04001570 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001571 if (retval)
1572 goto fput_fail;
1573 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001574
Jan Blunck5dd784d02008-02-14 19:34:38 -08001575 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001576 if (flags & LOOKUP_RCU) {
1577 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001578 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001579 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1580 br_read_lock(vfsmount_lock);
1581 rcu_read_lock();
1582 } else {
1583 path_get(&file->f_path);
1584 fput_light(file, fput_needed);
1585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 }
Al Viroe41f7d42011-02-22 14:02:58 -05001587
Nick Piggin31e6b012011-01-07 17:49:52 +11001588 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001589 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001590
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001591fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001592 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001593out_fail:
1594 return retval;
1595}
1596
Al Virobd92d7f2011-03-14 19:54:59 -04001597static inline int lookup_last(struct nameidata *nd, struct path *path)
1598{
1599 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1600 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1601
1602 nd->flags &= ~LOOKUP_PARENT;
1603 return walk_component(nd, path, &nd->last, nd->last_type,
1604 nd->flags & LOOKUP_FOLLOW);
1605}
1606
Al Viro9b4a9b12009-04-07 11:44:16 -04001607/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001608static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001609 unsigned int flags, struct nameidata *nd)
1610{
Al Viro70e9b352011-03-05 21:12:22 -05001611 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001612 struct path path;
1613 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001614
1615 /*
1616 * Path walking is largely split up into 2 different synchronisation
1617 * schemes, rcu-walk and ref-walk (explained in
1618 * Documentation/filesystems/path-lookup.txt). These share much of the
1619 * path walk code, but some things particularly setup, cleanup, and
1620 * following mounts are sufficiently divergent that functions are
1621 * duplicated. Typically there is a function foo(), and its RCU
1622 * analogue, foo_rcu().
1623 *
1624 * -ECHILD is the error number of choice (just to avoid clashes) that
1625 * is returned if some aspect of an rcu-walk fails. Such an error must
1626 * be handled by restarting a traditional ref-walk (which will always
1627 * be able to complete).
1628 */
Al Virobd92d7f2011-03-14 19:54:59 -04001629 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001630
Al Virobd92d7f2011-03-14 19:54:59 -04001631 if (unlikely(err))
1632 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001633
1634 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001635 err = link_path_walk(name, nd);
1636
1637 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001638 err = lookup_last(nd, &path);
1639 while (err > 0) {
1640 void *cookie;
1641 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001642 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001643 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001644 if (!err)
1645 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001646 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001647 }
1648 }
Al Viroee0827c2011-02-21 23:38:09 -05001649
Al Viro9f1fafe2011-03-25 11:00:12 -04001650 if (!err)
1651 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001652
1653 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1654 if (!nd->inode->i_op->lookup) {
1655 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001656 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001657 }
1658 }
Al Viro16c2cd72011-02-22 15:50:10 -05001659
Al Viro70e9b352011-03-05 21:12:22 -05001660 if (base)
1661 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001662
Al Viro5b6ca022011-03-09 23:04:47 -05001663 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001664 path_put(&nd->root);
1665 nd->root.mnt = NULL;
1666 }
Al Virobd92d7f2011-03-14 19:54:59 -04001667 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001668}
Nick Piggin31e6b012011-01-07 17:49:52 +11001669
Al Viroee0827c2011-02-21 23:38:09 -05001670static int do_path_lookup(int dfd, const char *name,
1671 unsigned int flags, struct nameidata *nd)
1672{
1673 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1674 if (unlikely(retval == -ECHILD))
1675 retval = path_lookupat(dfd, name, flags, nd);
1676 if (unlikely(retval == -ESTALE))
1677 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001678
1679 if (likely(!retval)) {
1680 if (unlikely(!audit_dummy_context())) {
1681 if (nd->path.dentry && nd->inode)
1682 audit_inode(name, nd->path.dentry);
1683 }
1684 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001685 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686}
1687
Al Viroc9c6cac2011-02-16 15:15:47 -05001688int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001689{
Al Viroc9c6cac2011-02-16 15:15:47 -05001690 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001691}
1692
Al Virod1811462008-08-02 00:49:18 -04001693int kern_path(const char *name, unsigned int flags, struct path *path)
1694{
1695 struct nameidata nd;
1696 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1697 if (!res)
1698 *path = nd.path;
1699 return res;
1700}
1701
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001702/**
1703 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1704 * @dentry: pointer to dentry of the base directory
1705 * @mnt: pointer to vfs mount of the base directory
1706 * @name: pointer to file name
1707 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001708 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001709 */
1710int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1711 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001712 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001713{
Al Viroe0a01242011-06-27 17:00:37 -04001714 struct nameidata nd;
1715 int err;
1716 nd.root.dentry = dentry;
1717 nd.root.mnt = mnt;
1718 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001719 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001720 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1721 if (!err)
1722 *path = nd.path;
1723 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001724}
1725
Christoph Hellwigeead1912007-10-16 23:25:38 -07001726static struct dentry *__lookup_hash(struct qstr *name,
1727 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001729 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001730 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 int err;
1732
Al Viro4ad5abb2011-06-20 19:57:03 -04001733 err = inode_permission(inode, MAY_EXEC);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001734 if (err)
1735 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
1737 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001738 * Don't bother with __d_lookup: callers are for creat as
1739 * well as unlink, so a lot of the time it would cost
1740 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001741 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001742 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001743
Josef Bacik44396f42011-05-31 11:58:49 -04001744 if (dentry && d_need_lookup(dentry)) {
1745 /*
1746 * __lookup_hash is called with the parent dir's i_mutex already
1747 * held, so we are good to go here.
1748 */
1749 dentry = d_inode_lookup(base, dentry, nd);
1750 if (IS_ERR(dentry))
1751 return dentry;
1752 }
1753
Al Virod2d9e9f2011-06-20 10:55:26 -04001754 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1755 int status = d_revalidate(dentry, nd);
1756 if (unlikely(status <= 0)) {
1757 /*
1758 * The dentry failed validation.
1759 * If d_revalidate returned 0 attempt to invalidate
1760 * the dentry otherwise d_revalidate is asking us
1761 * to return a fail status.
1762 */
1763 if (status < 0) {
1764 dput(dentry);
1765 return ERR_PTR(status);
1766 } else if (!d_invalidate(dentry)) {
1767 dput(dentry);
1768 dentry = NULL;
1769 }
1770 }
1771 }
Al Viro6e6b1bd2009-08-13 23:38:37 +04001772
Nick Pigginbaa03892010-08-18 04:37:31 +10001773 if (!dentry)
1774 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001775
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 return dentry;
1777}
1778
James Morris057f6c02007-04-26 00:12:05 -07001779/*
1780 * Restricted form of lookup. Doesn't follow links, single-component only,
1781 * needs parent already locked. Doesn't follow mounts.
1782 * SMP-safe.
1783 */
Adrian Bunka244e162006-03-31 02:32:11 -08001784static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
Jan Blunck4ac91372008-02-14 19:34:32 -08001786 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787}
1788
Christoph Hellwigeead1912007-10-16 23:25:38 -07001789/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001790 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001791 * @name: pathname component to lookup
1792 * @base: base directory to lookup from
1793 * @len: maximum length @len should be interpreted to
1794 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001795 * Note that this routine is purely a helper for filesystem usage and should
1796 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001797 * nameidata argument is passed to the filesystem methods and a filesystem
1798 * using this helper needs to be prepared for that.
1799 */
James Morris057f6c02007-04-26 00:12:05 -07001800struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1801{
James Morris057f6c02007-04-26 00:12:05 -07001802 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001803 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001804
David Woodhouse2f9092e2009-04-20 23:18:37 +01001805 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1806
Al Viro6a96ba52011-03-07 23:49:20 -05001807 this.name = name;
1808 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08001809 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05001810 if (!len)
1811 return ERR_PTR(-EACCES);
1812
Al Viro6a96ba52011-03-07 23:49:20 -05001813 while (len--) {
1814 c = *(const unsigned char *)name++;
1815 if (c == '/' || c == '\0')
1816 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05001817 }
Al Viro5a202bc2011-03-08 14:17:44 -05001818 /*
1819 * See if the low-level filesystem might want
1820 * to use its own hash..
1821 */
1822 if (base->d_flags & DCACHE_OP_HASH) {
1823 int err = base->d_op->d_hash(base, base->d_inode, &this);
1824 if (err < 0)
1825 return ERR_PTR(err);
1826 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001827
Christoph Hellwig49705b72005-11-08 21:35:06 -08001828 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001829}
1830
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001831int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1832 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833{
Al Viro2d8f3032008-07-22 09:59:21 -04001834 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001835 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001838
1839 BUG_ON(flags & LOOKUP_PARENT);
1840
1841 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001843 if (!err)
1844 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 }
1846 return err;
1847}
1848
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001849int user_path_at(int dfd, const char __user *name, unsigned flags,
1850 struct path *path)
1851{
1852 return user_path_at_empty(dfd, name, flags, path, 0);
1853}
1854
Al Viro2ad94ae2008-07-21 09:32:51 -04001855static int user_path_parent(int dfd, const char __user *path,
1856 struct nameidata *nd, char **name)
1857{
1858 char *s = getname(path);
1859 int error;
1860
1861 if (IS_ERR(s))
1862 return PTR_ERR(s);
1863
1864 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1865 if (error)
1866 putname(s);
1867 else
1868 *name = s;
1869
1870 return error;
1871}
1872
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873/*
1874 * It's inline, so penalty for filesystems that don't use sticky bit is
1875 * minimal.
1876 */
1877static inline int check_sticky(struct inode *dir, struct inode *inode)
1878{
David Howellsda9592e2008-11-14 10:39:05 +11001879 uid_t fsuid = current_fsuid();
1880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (!(dir->i_mode & S_ISVTX))
1882 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001883 if (current_user_ns() != inode_userns(inode))
1884 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001885 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001887 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001889
1890other_userns:
1891 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892}
1893
1894/*
1895 * Check whether we can remove a link victim from directory dir, check
1896 * whether the type of victim is right.
1897 * 1. We can't do it if dir is read-only (done in permission())
1898 * 2. We should have write and exec permissions on dir
1899 * 3. We can't remove anything from append-only dir
1900 * 4. We can't do anything with immutable dir (done in permission())
1901 * 5. If the sticky bit on dir is set we should either
1902 * a. be owner of dir, or
1903 * b. be owner of victim, or
1904 * c. have CAP_FOWNER capability
1905 * 6. If the victim is append-only or immutable we can't do antyhing with
1906 * links pointing to it.
1907 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1908 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1909 * 9. We can't remove a root or mountpoint.
1910 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1911 * nfs_async_unlink().
1912 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001913static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914{
1915 int error;
1916
1917 if (!victim->d_inode)
1918 return -ENOENT;
1919
1920 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001921 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
Al Virof419a2e2008-07-22 00:07:17 -04001923 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 if (error)
1925 return error;
1926 if (IS_APPEND(dir))
1927 return -EPERM;
1928 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001929 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 return -EPERM;
1931 if (isdir) {
1932 if (!S_ISDIR(victim->d_inode->i_mode))
1933 return -ENOTDIR;
1934 if (IS_ROOT(victim))
1935 return -EBUSY;
1936 } else if (S_ISDIR(victim->d_inode->i_mode))
1937 return -EISDIR;
1938 if (IS_DEADDIR(dir))
1939 return -ENOENT;
1940 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1941 return -EBUSY;
1942 return 0;
1943}
1944
1945/* Check whether we can create an object with dentry child in directory
1946 * dir.
1947 * 1. We can't do it if child already exists (open has special treatment for
1948 * this case, but since we are inlined it's OK)
1949 * 2. We can't do it if dir is read-only (done in permission())
1950 * 3. We should have write and exec permissions on dir
1951 * 4. We can't do it if dir is immutable (done in permission())
1952 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001953static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954{
1955 if (child->d_inode)
1956 return -EEXIST;
1957 if (IS_DEADDIR(dir))
1958 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001959 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960}
1961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962/*
1963 * p1 and p2 should be directories on the same fs.
1964 */
1965struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1966{
1967 struct dentry *p;
1968
1969 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001970 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 return NULL;
1972 }
1973
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001974 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001976 p = d_ancestor(p2, p1);
1977 if (p) {
1978 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1979 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1980 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
1982
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001983 p = d_ancestor(p1, p2);
1984 if (p) {
1985 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1986 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1987 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 }
1989
Ingo Molnarf2eace22006-07-03 00:25:05 -07001990 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1991 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return NULL;
1993}
1994
1995void unlock_rename(struct dentry *p1, struct dentry *p2)
1996{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001997 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001999 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002000 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 }
2002}
2003
Al Viro4acdaf22011-07-26 01:42:34 -04002004int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 struct nameidata *nd)
2006{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002007 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
2009 if (error)
2010 return error;
2011
Al Viroacfa4382008-12-04 10:06:33 -05002012 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 return -EACCES; /* shouldn't it be ENOSYS? */
2014 mode &= S_IALLUGO;
2015 mode |= S_IFREG;
2016 error = security_inode_create(dir, dentry, mode);
2017 if (error)
2018 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002020 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002021 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 return error;
2023}
2024
Al Viro73d049a2011-03-11 12:08:24 -05002025static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002027 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 struct inode *inode = dentry->d_inode;
2029 int error;
2030
Al Virobcda7652011-03-13 16:42:14 -04002031 /* O_PATH? */
2032 if (!acc_mode)
2033 return 0;
2034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 if (!inode)
2036 return -ENOENT;
2037
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002038 switch (inode->i_mode & S_IFMT) {
2039 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002041 case S_IFDIR:
2042 if (acc_mode & MAY_WRITE)
2043 return -EISDIR;
2044 break;
2045 case S_IFBLK:
2046 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002047 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002049 /*FALLTHRU*/
2050 case S_IFIFO:
2051 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002053 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002054 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002055
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002056 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002057 if (error)
2058 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 /*
2061 * An append-only file must be opened in append mode for writing.
2062 */
2063 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002064 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002065 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002067 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 }
2069
2070 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002071 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002072 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002074 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002075}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
Jeff Laytone1181ee2010-12-07 16:19:50 -05002077static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002078{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002079 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002080 struct inode *inode = path->dentry->d_inode;
2081 int error = get_write_access(inode);
2082 if (error)
2083 return error;
2084 /*
2085 * Refuse to truncate files with mandatory locks held on them.
2086 */
2087 error = locks_verify_locked(inode);
2088 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002089 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002090 if (!error) {
2091 error = do_truncate(path->dentry, 0,
2092 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002093 filp);
Al Viro7715b522009-12-16 03:54:00 -05002094 }
2095 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002096 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097}
2098
Dave Hansend57999e2008-02-15 14:37:27 -08002099static inline int open_to_namei_flags(int flag)
2100{
Al Viro8a5e9292011-06-25 19:15:54 -04002101 if ((flag & O_ACCMODE) == 3)
2102 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002103 return flag;
2104}
2105
Nick Piggin31e6b012011-01-07 17:49:52 +11002106/*
Al Virofe2d35f2011-03-05 22:58:25 -05002107 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002108 */
Al Virofb1cc552009-12-24 01:58:28 -05002109static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002110 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002111{
Al Viroa1e28032009-12-24 02:12:06 -05002112 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002113 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002114 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002115 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002116 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002117 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002118 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002119 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002120
Al Viroc3e380b2011-02-23 13:39:45 -05002121 nd->flags &= ~LOOKUP_PARENT;
2122 nd->flags |= op->intent;
2123
Al Viro1f36f772009-12-26 10:56:19 -05002124 switch (nd->last_type) {
2125 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002126 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002127 error = handle_dots(nd, nd->last_type);
2128 if (error)
2129 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002130 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002131 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002132 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002133 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002134 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002135 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002136 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002137 error = -EISDIR;
2138 goto exit;
2139 }
2140 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002141 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002142 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002143 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002144 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002145 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002146 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002147 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002148
Al Viroca344a892011-03-09 00:36:45 -05002149 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002150 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002151 if (nd->last.name[nd->last.len])
2152 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002153 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2154 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002155 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002156 error = walk_component(nd, path, &nd->last, LAST_NORM,
2157 !symlink_ok);
2158 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002159 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002160 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002161 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002162 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002163 error = complete_walk(nd);
2164 if (error)
Miklos Szeredi7f6c7e62012-03-06 13:56:34 +01002165 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002166
2167 error = -ENOTDIR;
2168 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002169 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002170 goto exit;
2171 }
2172 audit_inode(pathname, nd->path.dentry);
2173 goto ok;
2174 }
2175
2176 /* create side of things */
Al Viroa3fbbde2011-11-07 21:21:26 +00002177 /*
2178 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2179 * cleared when we got to the last component we are about to look up
2180 */
Al Viro9f1fafe2011-03-25 11:00:12 -04002181 error = complete_walk(nd);
2182 if (error)
2183 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002184
2185 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002186 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002187 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002188 if (nd->last.name[nd->last.len])
2189 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002190
Al Viroa1e28032009-12-24 02:12:06 -05002191 mutex_lock(&dir->d_inode->i_mutex);
2192
Al Viro6c0d46c2011-03-09 00:59:59 -05002193 dentry = lookup_hash(nd);
2194 error = PTR_ERR(dentry);
2195 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002196 mutex_unlock(&dir->d_inode->i_mutex);
2197 goto exit;
2198 }
2199
Al Viro6c0d46c2011-03-09 00:59:59 -05002200 path->dentry = dentry;
2201 path->mnt = nd->path.mnt;
2202
Al Virofb1cc552009-12-24 01:58:28 -05002203 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002204 if (!dentry->d_inode) {
Al Viroa218d0f2011-11-21 14:59:34 -05002205 umode_t mode = op->mode;
Al Viro6c0d46c2011-03-09 00:59:59 -05002206 if (!IS_POSIXACL(dir->d_inode))
2207 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002208 /*
2209 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002210 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002211 * the time when the file is created and when
2212 * a permanent write count is taken through
2213 * the 'struct file' in nameidata_to_filp().
2214 */
2215 error = mnt_want_write(nd->path.mnt);
2216 if (error)
2217 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002218 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002219 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002220 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002221 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002222 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002223 error = security_path_mknod(&nd->path, dentry, mode, 0);
2224 if (error)
2225 goto exit_mutex_unlock;
2226 error = vfs_create(dir->d_inode, dentry, mode, nd);
2227 if (error)
2228 goto exit_mutex_unlock;
2229 mutex_unlock(&dir->d_inode->i_mutex);
2230 dput(nd->path.dentry);
2231 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002232 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002233 }
2234
2235 /*
2236 * It already exists.
2237 */
2238 mutex_unlock(&dir->d_inode->i_mutex);
2239 audit_inode(pathname, path->dentry);
2240
2241 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002242 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002243 goto exit_dput;
2244
David Howells9875cf82011-01-14 18:45:21 +00002245 error = follow_managed(path, nd->flags);
2246 if (error < 0)
2247 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002248
Al Viroa3fbbde2011-11-07 21:21:26 +00002249 if (error)
2250 nd->flags |= LOOKUP_JUMPED;
2251
Al Virofb1cc552009-12-24 01:58:28 -05002252 error = -ENOENT;
2253 if (!path->dentry->d_inode)
2254 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002255
2256 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002257 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002258
2259 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002260 nd->inode = path->dentry->d_inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002261 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2262 error = complete_walk(nd);
2263 if (error)
Miklos Szeredi097b1802012-03-06 13:56:33 +01002264 return ERR_PTR(error);
Al Virofb1cc552009-12-24 01:58:28 -05002265 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002266 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002267 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002268ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002269 if (!S_ISREG(nd->inode->i_mode))
2270 will_truncate = 0;
2271
Al Viro0f9d1a12011-03-09 00:13:14 -05002272 if (will_truncate) {
2273 error = mnt_want_write(nd->path.mnt);
2274 if (error)
2275 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002276 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002277 }
Al Viroca344a892011-03-09 00:36:45 -05002278common:
Al Virobcda7652011-03-13 16:42:14 -04002279 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002280 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002281 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002282 filp = nameidata_to_filp(nd);
2283 if (!IS_ERR(filp)) {
2284 error = ima_file_check(filp, op->acc_mode);
2285 if (error) {
2286 fput(filp);
2287 filp = ERR_PTR(error);
2288 }
2289 }
2290 if (!IS_ERR(filp)) {
2291 if (will_truncate) {
2292 error = handle_truncate(filp);
2293 if (error) {
2294 fput(filp);
2295 filp = ERR_PTR(error);
2296 }
2297 }
2298 }
Al Viroca344a892011-03-09 00:36:45 -05002299out:
2300 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002301 mnt_drop_write(nd->path.mnt);
2302 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002303 return filp;
2304
2305exit_mutex_unlock:
2306 mutex_unlock(&dir->d_inode->i_mutex);
2307exit_dput:
2308 path_put_conditional(path, nd);
2309exit:
Al Viroca344a892011-03-09 00:36:45 -05002310 filp = ERR_PTR(error);
2311 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002312}
2313
Al Viro13aab422011-02-23 17:54:08 -05002314static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002315 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316{
Al Virofe2d35f2011-03-05 22:58:25 -05002317 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002318 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002319 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002320 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002321
2322 filp = get_empty_filp();
2323 if (!filp)
2324 return ERR_PTR(-ENFILE);
2325
Al Viro47c805d2011-02-23 17:44:09 -05002326 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002327 nd->intent.open.file = filp;
2328 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2329 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002330
Al Viro73d049a2011-03-11 12:08:24 -05002331 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002332 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002333 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002334
Al Virofe2d35f2011-03-05 22:58:25 -05002335 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002336 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002337 if (unlikely(error))
2338 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Al Viro73d049a2011-03-11 12:08:24 -05002340 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002341 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002342 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002343 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002344 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002345 path_put_conditional(&path, nd);
2346 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002347 filp = ERR_PTR(-ELOOP);
2348 break;
2349 }
Al Viro73d049a2011-03-11 12:08:24 -05002350 nd->flags |= LOOKUP_PARENT;
2351 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002352 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002353 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002354 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002355 else
Al Viro73d049a2011-03-11 12:08:24 -05002356 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002357 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002358 }
Al Viro10fa8e62009-12-26 07:09:49 -05002359out:
Al Viro73d049a2011-03-11 12:08:24 -05002360 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2361 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002362 if (base)
2363 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002364 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002365 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Nick Piggin31e6b012011-01-07 17:49:52 +11002367out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002368 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002369 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370}
2371
Al Viro13aab422011-02-23 17:54:08 -05002372struct file *do_filp_open(int dfd, const char *pathname,
2373 const struct open_flags *op, int flags)
2374{
Al Viro73d049a2011-03-11 12:08:24 -05002375 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002376 struct file *filp;
2377
Al Viro73d049a2011-03-11 12:08:24 -05002378 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002379 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002380 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002381 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002382 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002383 return filp;
2384}
2385
Al Viro73d049a2011-03-11 12:08:24 -05002386struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2387 const char *name, const struct open_flags *op, int flags)
2388{
2389 struct nameidata nd;
2390 struct file *file;
2391
2392 nd.root.mnt = mnt;
2393 nd.root.dentry = dentry;
2394
2395 flags |= LOOKUP_ROOT;
2396
Al Virobcda7652011-03-13 16:42:14 -04002397 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002398 return ERR_PTR(-ELOOP);
2399
2400 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2401 if (unlikely(file == ERR_PTR(-ECHILD)))
2402 file = path_openat(-1, name, &nd, op, flags);
2403 if (unlikely(file == ERR_PTR(-ESTALE)))
2404 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2405 return file;
2406}
2407
Al Viroed75e952011-06-27 16:53:43 -04002408struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002410 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002411 struct nameidata nd;
2412 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2413 if (error)
2414 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002416 /*
2417 * Yucky last component or no last component at all?
2418 * (foo/., foo/.., /////)
2419 */
Al Viroed75e952011-06-27 16:53:43 -04002420 if (nd.last_type != LAST_NORM)
2421 goto out;
2422 nd.flags &= ~LOOKUP_PARENT;
2423 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2424 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002425
2426 /*
2427 * Do the final lookup.
2428 */
Al Viroed75e952011-06-27 16:53:43 -04002429 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2430 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 if (IS_ERR(dentry))
2432 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002433
Al Viroe9baf6e2008-05-15 04:49:12 -04002434 if (dentry->d_inode)
2435 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002436 /*
2437 * Special case - lookup gave negative, but... we had foo/bar/
2438 * From the vfs_mknod() POV we just have a negative dentry -
2439 * all is fine. Let's be bastards - you had / on the end, you've
2440 * been asking for (non-existent) directory. -ENOENT for you.
2441 */
Al Viroed75e952011-06-27 16:53:43 -04002442 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002443 dput(dentry);
2444 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002445 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002446 }
Al Viroed75e952011-06-27 16:53:43 -04002447 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002449eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002451 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452fail:
Al Viroed75e952011-06-27 16:53:43 -04002453 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2454out:
2455 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 return dentry;
2457}
Al Virodae6ad82011-06-26 11:50:15 -04002458EXPORT_SYMBOL(kern_path_create);
2459
2460struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2461{
2462 char *tmp = getname(pathname);
2463 struct dentry *res;
2464 if (IS_ERR(tmp))
2465 return ERR_CAST(tmp);
2466 res = kern_path_create(dfd, tmp, path, is_dir);
2467 putname(tmp);
2468 return res;
2469}
2470EXPORT_SYMBOL(user_path_create);
2471
Al Viro1a67aaf2011-07-26 01:52:52 -04002472int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002474 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
2476 if (error)
2477 return error;
2478
Serge E. Hallyne795b712011-03-23 16:43:25 -07002479 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2480 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 return -EPERM;
2482
Al Viroacfa4382008-12-04 10:06:33 -05002483 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 return -EPERM;
2485
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002486 error = devcgroup_inode_mknod(mode, dev);
2487 if (error)
2488 return error;
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 error = security_inode_mknod(dir, dentry, mode, dev);
2491 if (error)
2492 return error;
2493
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002495 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002496 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 return error;
2498}
2499
Al Virof69aac02011-07-26 04:31:40 -04002500static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08002501{
2502 switch (mode & S_IFMT) {
2503 case S_IFREG:
2504 case S_IFCHR:
2505 case S_IFBLK:
2506 case S_IFIFO:
2507 case S_IFSOCK:
2508 case 0: /* zero mode translates to S_IFREG */
2509 return 0;
2510 case S_IFDIR:
2511 return -EPERM;
2512 default:
2513 return -EINVAL;
2514 }
2515}
2516
Al Viro8208a222011-07-25 17:32:17 -04002517SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002518 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519{
Al Viro2ad94ae2008-07-21 09:32:51 -04002520 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002521 struct path path;
2522 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
2524 if (S_ISDIR(mode))
2525 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Al Virodae6ad82011-06-26 11:50:15 -04002527 dentry = user_path_create(dfd, filename, &path, 0);
2528 if (IS_ERR(dentry))
2529 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002530
Al Virodae6ad82011-06-26 11:50:15 -04002531 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002532 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002533 error = may_mknod(mode);
2534 if (error)
2535 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002536 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002537 if (error)
2538 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002539 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002540 if (error)
2541 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002542 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002544 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 break;
2546 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002547 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 new_decode_dev(dev));
2549 break;
2550 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002551 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002554out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002555 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002556out_dput:
2557 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002558 mutex_unlock(&path.dentry->d_inode->i_mutex);
2559 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
2561 return error;
2562}
2563
Al Viro8208a222011-07-25 17:32:17 -04002564SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002565{
2566 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2567}
2568
Al Viro18bb1db2011-07-26 01:41:39 -04002569int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002571 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
2573 if (error)
2574 return error;
2575
Al Viroacfa4382008-12-04 10:06:33 -05002576 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 return -EPERM;
2578
2579 mode &= (S_IRWXUGO|S_ISVTX);
2580 error = security_inode_mkdir(dir, dentry, mode);
2581 if (error)
2582 return error;
2583
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002585 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002586 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 return error;
2588}
2589
Al Viroa218d0f2011-11-21 14:59:34 -05002590SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591{
Dave Hansen6902d922006-09-30 23:29:01 -07002592 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002593 struct path path;
2594 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
Al Virodae6ad82011-06-26 11:50:15 -04002596 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002597 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002598 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002599
Al Virodae6ad82011-06-26 11:50:15 -04002600 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002601 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002602 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002603 if (error)
2604 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002605 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002606 if (error)
2607 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002608 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002609out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002610 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002611out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002612 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002613 mutex_unlock(&path.dentry->d_inode->i_mutex);
2614 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 return error;
2616}
2617
Al Viroa218d0f2011-11-21 14:59:34 -05002618SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002619{
2620 return sys_mkdirat(AT_FDCWD, pathname, mode);
2621}
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623/*
Sage Weila71905f2011-05-24 13:06:08 -07002624 * The dentry_unhash() helper will try to drop the dentry early: we
2625 * should have a usage count of 2 if we're the only user of this
2626 * dentry, and if that is true (possibly after pruning the dcache),
2627 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 *
2629 * A low-level filesystem can, if it choses, legally
2630 * do a
2631 *
2632 * if (!d_unhashed(dentry))
2633 * return -EBUSY;
2634 *
2635 * if it cannot handle the case of removing a directory
2636 * that is still in use by something else..
2637 */
2638void dentry_unhash(struct dentry *dentry)
2639{
Vasily Averindc168422006-12-06 20:37:07 -08002640 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002642 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 __d_drop(dentry);
2644 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645}
2646
2647int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2648{
2649 int error = may_delete(dir, dentry, 1);
2650
2651 if (error)
2652 return error;
2653
Al Viroacfa4382008-12-04 10:06:33 -05002654 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 return -EPERM;
2656
Al Viro1d2ef592011-09-14 18:55:41 +01002657 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002658 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
Sage Weil912dbc12011-05-24 13:06:11 -07002660 error = -EBUSY;
2661 if (d_mountpoint(dentry))
2662 goto out;
2663
2664 error = security_inode_rmdir(dir, dentry);
2665 if (error)
2666 goto out;
2667
Sage Weil3cebde22011-05-29 21:20:59 -07002668 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002669 error = dir->i_op->rmdir(dir, dentry);
2670 if (error)
2671 goto out;
2672
2673 dentry->d_inode->i_flags |= S_DEAD;
2674 dont_mount(dentry);
2675
2676out:
2677 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01002678 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002679 if (!error)
2680 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 return error;
2682}
2683
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002684static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685{
2686 int error = 0;
2687 char * name;
2688 struct dentry *dentry;
2689 struct nameidata nd;
2690
Al Viro2ad94ae2008-07-21 09:32:51 -04002691 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002693 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
2695 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002696 case LAST_DOTDOT:
2697 error = -ENOTEMPTY;
2698 goto exit1;
2699 case LAST_DOT:
2700 error = -EINVAL;
2701 goto exit1;
2702 case LAST_ROOT:
2703 error = -EBUSY;
2704 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002706
2707 nd.flags &= ~LOOKUP_PARENT;
2708
Jan Blunck4ac91372008-02-14 19:34:32 -08002709 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002710 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002712 if (IS_ERR(dentry))
2713 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002714 if (!dentry->d_inode) {
2715 error = -ENOENT;
2716 goto exit3;
2717 }
Dave Hansen06227532008-02-15 14:37:34 -08002718 error = mnt_want_write(nd.path.mnt);
2719 if (error)
2720 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002721 error = security_path_rmdir(&nd.path, dentry);
2722 if (error)
2723 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002724 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002725exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002726 mnt_drop_write(nd.path.mnt);
2727exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002728 dput(dentry);
2729exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002730 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002732 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 putname(name);
2734 return error;
2735}
2736
Heiko Carstens3cdad422009-01-14 14:14:22 +01002737SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002738{
2739 return do_rmdir(AT_FDCWD, pathname);
2740}
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742int vfs_unlink(struct inode *dir, struct dentry *dentry)
2743{
2744 int error = may_delete(dir, dentry, 0);
2745
2746 if (error)
2747 return error;
2748
Al Viroacfa4382008-12-04 10:06:33 -05002749 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 return -EPERM;
2751
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002752 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 if (d_mountpoint(dentry))
2754 error = -EBUSY;
2755 else {
2756 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002757 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002759 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002760 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002763 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
2765 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2766 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002767 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 }
Robert Love0eeca282005-07-12 17:06:03 -04002770
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 return error;
2772}
2773
2774/*
2775 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002776 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 * writeout happening, and we don't want to prevent access to the directory
2778 * while waiting on the I/O.
2779 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002780static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781{
Al Viro2ad94ae2008-07-21 09:32:51 -04002782 int error;
2783 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 struct dentry *dentry;
2785 struct nameidata nd;
2786 struct inode *inode = NULL;
2787
Al Viro2ad94ae2008-07-21 09:32:51 -04002788 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002790 return error;
2791
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 error = -EISDIR;
2793 if (nd.last_type != LAST_NORM)
2794 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002795
2796 nd.flags &= ~LOOKUP_PARENT;
2797
Jan Blunck4ac91372008-02-14 19:34:32 -08002798 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002799 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 error = PTR_ERR(dentry);
2801 if (!IS_ERR(dentry)) {
2802 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002803 if (nd.last.name[nd.last.len])
2804 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002806 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002807 goto slashes;
2808 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002809 error = mnt_want_write(nd.path.mnt);
2810 if (error)
2811 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002812 error = security_path_unlink(&nd.path, dentry);
2813 if (error)
2814 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002815 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002816exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002817 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 exit2:
2819 dput(dentry);
2820 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002821 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 if (inode)
2823 iput(inode); /* truncate the inode here */
2824exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002825 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 putname(name);
2827 return error;
2828
2829slashes:
2830 error = !dentry->d_inode ? -ENOENT :
2831 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2832 goto exit2;
2833}
2834
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002835SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002836{
2837 if ((flag & ~AT_REMOVEDIR) != 0)
2838 return -EINVAL;
2839
2840 if (flag & AT_REMOVEDIR)
2841 return do_rmdir(dfd, pathname);
2842
2843 return do_unlinkat(dfd, pathname);
2844}
2845
Heiko Carstens3480b252009-01-14 14:14:16 +01002846SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002847{
2848 return do_unlinkat(AT_FDCWD, pathname);
2849}
2850
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002851int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002853 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
2855 if (error)
2856 return error;
2857
Al Viroacfa4382008-12-04 10:06:33 -05002858 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 return -EPERM;
2860
2861 error = security_inode_symlink(dir, dentry, oldname);
2862 if (error)
2863 return error;
2864
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002866 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002867 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 return error;
2869}
2870
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002871SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2872 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873{
Al Viro2ad94ae2008-07-21 09:32:51 -04002874 int error;
2875 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07002876 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002877 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878
2879 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002880 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002882
Al Virodae6ad82011-06-26 11:50:15 -04002883 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07002884 error = PTR_ERR(dentry);
2885 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002886 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07002887
Al Virodae6ad82011-06-26 11:50:15 -04002888 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002889 if (error)
2890 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002891 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002892 if (error)
2893 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002894 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002895out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002896 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002897out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002898 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002899 mutex_unlock(&path.dentry->d_inode->i_mutex);
2900 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07002901out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 putname(from);
2903 return error;
2904}
2905
Heiko Carstens3480b252009-01-14 14:14:16 +01002906SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002907{
2908 return sys_symlinkat(oldname, AT_FDCWD, newname);
2909}
2910
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2912{
2913 struct inode *inode = old_dentry->d_inode;
2914 int error;
2915
2916 if (!inode)
2917 return -ENOENT;
2918
Miklos Szeredia95164d2008-07-30 15:08:48 +02002919 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 if (error)
2921 return error;
2922
2923 if (dir->i_sb != inode->i_sb)
2924 return -EXDEV;
2925
2926 /*
2927 * A link to an append-only or immutable file cannot be created.
2928 */
2929 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2930 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002931 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002933 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 return -EPERM;
2935
2936 error = security_inode_link(old_dentry, dir, new_dentry);
2937 if (error)
2938 return error;
2939
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002940 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302941 /* Make sure we don't allow creating hardlink to an unlinked file */
2942 if (inode->i_nlink == 0)
2943 error = -ENOENT;
2944 else
2945 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002946 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002947 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002948 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 return error;
2950}
2951
2952/*
2953 * Hardlinks are often used in delicate situations. We avoid
2954 * security-related surprises by not following symlinks on the
2955 * newname. --KAB
2956 *
2957 * We don't follow them on the oldname either to be compatible
2958 * with linux 2.0, and to avoid hard-linking to directories
2959 * and other special files. --ADM
2960 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002961SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2962 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963{
2964 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002965 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302966 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302969 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002970 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302971 /*
2972 * To use null names we require CAP_DAC_READ_SEARCH
2973 * This ensures that not everyone will be able to create
2974 * handlink using the passed filedescriptor.
2975 */
2976 if (flags & AT_EMPTY_PATH) {
2977 if (!capable(CAP_DAC_READ_SEARCH))
2978 return -ENOENT;
2979 how = LOOKUP_EMPTY;
2980 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002981
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302982 if (flags & AT_SYMLINK_FOLLOW)
2983 how |= LOOKUP_FOLLOW;
2984
2985 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002987 return error;
2988
Al Virodae6ad82011-06-26 11:50:15 -04002989 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002991 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002992 goto out;
2993
2994 error = -EXDEV;
2995 if (old_path.mnt != new_path.mnt)
2996 goto out_dput;
2997 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002998 if (error)
2999 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003000 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003001 if (error)
3002 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003003 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003004out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003005 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003006out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003007 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003008 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3009 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010out:
Al Viro2d8f3032008-07-22 09:59:21 -04003011 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012
3013 return error;
3014}
3015
Heiko Carstens3480b252009-01-14 14:14:16 +01003016SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003017{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003018 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003019}
3020
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021/*
3022 * The worst of all namespace operations - renaming directory. "Perverted"
3023 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3024 * Problems:
3025 * a) we can get into loop creation. Check is done in is_subdir().
3026 * b) race potential - two innocent renames can create a loop together.
3027 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003028 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 * story.
3030 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003031 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 * whether the target exists). Solution: try to be smart with locking
3033 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003034 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 * move will be locked. Thus we can rank directories by the tree
3036 * (ancestors first) and rank all non-directories after them.
3037 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003038 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 * HOWEVER, it relies on the assumption that any object with ->lookup()
3040 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3041 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003042 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003043 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003045 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 * locking].
3047 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003048static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3049 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050{
3051 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003052 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
3054 /*
3055 * If we are going to change the parent - check write permissions,
3056 * we'll need to flip '..'.
3057 */
3058 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003059 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 if (error)
3061 return error;
3062 }
3063
3064 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3065 if (error)
3066 return error;
3067
Al Viro1d2ef592011-09-14 18:55:41 +01003068 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003069 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003070 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003071
3072 error = -EBUSY;
3073 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3074 goto out;
3075
Sage Weil3cebde22011-05-29 21:20:59 -07003076 if (target)
3077 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003078 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3079 if (error)
3080 goto out;
3081
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003083 target->i_flags |= S_DEAD;
3084 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 }
Sage Weil9055cba2011-05-24 13:06:12 -07003086out:
3087 if (target)
3088 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003089 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003090 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003091 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3092 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 return error;
3094}
3095
Adrian Bunk75c96f82005-05-05 16:16:09 -07003096static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3097 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098{
Sage Weil51892bb2011-05-24 13:06:13 -07003099 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 int error;
3101
3102 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3103 if (error)
3104 return error;
3105
3106 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003108 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003109
3110 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003112 goto out;
3113
3114 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3115 if (error)
3116 goto out;
3117
3118 if (target)
3119 dont_mount(new_dentry);
3120 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3121 d_move(old_dentry, new_dentry);
3122out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003124 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 dput(new_dentry);
3126 return error;
3127}
3128
3129int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3130 struct inode *new_dir, struct dentry *new_dentry)
3131{
3132 int error;
3133 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003134 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135
3136 if (old_dentry->d_inode == new_dentry->d_inode)
3137 return 0;
3138
3139 error = may_delete(old_dir, old_dentry, is_dir);
3140 if (error)
3141 return error;
3142
3143 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003144 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 else
3146 error = may_delete(new_dir, new_dentry, is_dir);
3147 if (error)
3148 return error;
3149
Al Viroacfa4382008-12-04 10:06:33 -05003150 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 return -EPERM;
3152
Robert Love0eeca282005-07-12 17:06:03 -04003153 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3154
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 if (is_dir)
3156 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3157 else
3158 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003159 if (!error)
3160 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003161 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003162 fsnotify_oldname_free(old_name);
3163
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 return error;
3165}
3166
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003167SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3168 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169{
Al Viro2ad94ae2008-07-21 09:32:51 -04003170 struct dentry *old_dir, *new_dir;
3171 struct dentry *old_dentry, *new_dentry;
3172 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003174 char *from;
3175 char *to;
3176 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177
Al Viro2ad94ae2008-07-21 09:32:51 -04003178 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 if (error)
3180 goto exit;
3181
Al Viro2ad94ae2008-07-21 09:32:51 -04003182 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 if (error)
3184 goto exit1;
3185
3186 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003187 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 goto exit2;
3189
Jan Blunck4ac91372008-02-14 19:34:32 -08003190 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 error = -EBUSY;
3192 if (oldnd.last_type != LAST_NORM)
3193 goto exit2;
3194
Jan Blunck4ac91372008-02-14 19:34:32 -08003195 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 if (newnd.last_type != LAST_NORM)
3197 goto exit2;
3198
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003199 oldnd.flags &= ~LOOKUP_PARENT;
3200 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003201 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003202
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 trap = lock_rename(new_dir, old_dir);
3204
Christoph Hellwig49705b72005-11-08 21:35:06 -08003205 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 error = PTR_ERR(old_dentry);
3207 if (IS_ERR(old_dentry))
3208 goto exit3;
3209 /* source must exist */
3210 error = -ENOENT;
3211 if (!old_dentry->d_inode)
3212 goto exit4;
3213 /* unless the source is a directory trailing slashes give -ENOTDIR */
3214 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3215 error = -ENOTDIR;
3216 if (oldnd.last.name[oldnd.last.len])
3217 goto exit4;
3218 if (newnd.last.name[newnd.last.len])
3219 goto exit4;
3220 }
3221 /* source should not be ancestor of target */
3222 error = -EINVAL;
3223 if (old_dentry == trap)
3224 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003225 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 error = PTR_ERR(new_dentry);
3227 if (IS_ERR(new_dentry))
3228 goto exit4;
3229 /* target should not be an ancestor of source */
3230 error = -ENOTEMPTY;
3231 if (new_dentry == trap)
3232 goto exit5;
3233
Dave Hansen9079b1e2008-02-15 14:37:49 -08003234 error = mnt_want_write(oldnd.path.mnt);
3235 if (error)
3236 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003237 error = security_path_rename(&oldnd.path, old_dentry,
3238 &newnd.path, new_dentry);
3239 if (error)
3240 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 error = vfs_rename(old_dir->d_inode, old_dentry,
3242 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003243exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003244 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245exit5:
3246 dput(new_dentry);
3247exit4:
3248 dput(old_dentry);
3249exit3:
3250 unlock_rename(new_dir, old_dir);
3251exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003252 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003253 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003255 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003257exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 return error;
3259}
3260
Heiko Carstensa26eab22009-01-14 14:14:17 +01003261SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003262{
3263 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3264}
3265
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3267{
3268 int len;
3269
3270 len = PTR_ERR(link);
3271 if (IS_ERR(link))
3272 goto out;
3273
3274 len = strlen(link);
3275 if (len > (unsigned) buflen)
3276 len = buflen;
3277 if (copy_to_user(buffer, link, len))
3278 len = -EFAULT;
3279out:
3280 return len;
3281}
3282
3283/*
3284 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3285 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3286 * using) it for any given inode is up to filesystem.
3287 */
3288int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3289{
3290 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003291 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003292 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003293
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003295 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003296 if (IS_ERR(cookie))
3297 return PTR_ERR(cookie);
3298
3299 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3300 if (dentry->d_inode->i_op->put_link)
3301 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3302 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303}
3304
3305int vfs_follow_link(struct nameidata *nd, const char *link)
3306{
3307 return __vfs_follow_link(nd, link);
3308}
3309
3310/* get the link contents into pagecache */
3311static char *page_getlink(struct dentry * dentry, struct page **ppage)
3312{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003313 char *kaddr;
3314 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003316 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003318 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003320 kaddr = kmap(page);
3321 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3322 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323}
3324
3325int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3326{
3327 struct page *page = NULL;
3328 char *s = page_getlink(dentry, &page);
3329 int res = vfs_readlink(dentry,buffer,buflen,s);
3330 if (page) {
3331 kunmap(page);
3332 page_cache_release(page);
3333 }
3334 return res;
3335}
3336
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003337void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003339 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003341 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342}
3343
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003344void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003346 struct page *page = cookie;
3347
3348 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 kunmap(page);
3350 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 }
3352}
3353
Nick Piggin54566b22009-01-04 12:00:53 -08003354/*
3355 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3356 */
3357int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358{
3359 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003360 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003361 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003362 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003364 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3365 if (nofs)
3366 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367
NeilBrown7e53cac2006-03-25 03:07:57 -08003368retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003369 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003370 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003372 goto fail;
3373
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 kaddr = kmap_atomic(page, KM_USER0);
3375 memcpy(kaddr, symname, len-1);
3376 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003377
3378 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3379 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 if (err < 0)
3381 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003382 if (err < len-1)
3383 goto retry;
3384
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 mark_inode_dirty(inode);
3386 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387fail:
3388 return err;
3389}
3390
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003391int page_symlink(struct inode *inode, const char *symname, int len)
3392{
3393 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003394 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003395}
3396
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003397const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 .readlink = generic_readlink,
3399 .follow_link = page_follow_link_light,
3400 .put_link = page_put_link,
3401};
3402
Al Viro2d8f3032008-07-22 09:59:21 -04003403EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003404EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405EXPORT_SYMBOL(follow_down);
3406EXPORT_SYMBOL(follow_up);
3407EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3408EXPORT_SYMBOL(getname);
3409EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410EXPORT_SYMBOL(lookup_one_len);
3411EXPORT_SYMBOL(page_follow_link_light);
3412EXPORT_SYMBOL(page_put_link);
3413EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003414EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415EXPORT_SYMBOL(page_symlink);
3416EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003417EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003418EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003419EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420EXPORT_SYMBOL(unlock_rename);
3421EXPORT_SYMBOL(vfs_create);
3422EXPORT_SYMBOL(vfs_follow_link);
3423EXPORT_SYMBOL(vfs_link);
3424EXPORT_SYMBOL(vfs_mkdir);
3425EXPORT_SYMBOL(vfs_mknod);
3426EXPORT_SYMBOL(generic_permission);
3427EXPORT_SYMBOL(vfs_readlink);
3428EXPORT_SYMBOL(vfs_rename);
3429EXPORT_SYMBOL(vfs_rmdir);
3430EXPORT_SYMBOL(vfs_symlink);
3431EXPORT_SYMBOL(vfs_unlink);
3432EXPORT_SYMBOL(dentry_unhash);
3433EXPORT_SYMBOL(generic_readlink);