blob: 0062dd17eb55d0f7bc365d0e06064cfb34d43449 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#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{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700164 return getname_flags(filename, 0, NULL);
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
Al Viro68ac1232012-03-15 08:21:57 -0400645 touch_atime(link);
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/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001057 * This looks up the name in dcache, possibly revalidates the old dentry and
1058 * allocates a new one if not found or not valid. In the need_lookup argument
1059 * returns whether i_op->lookup is necessary.
1060 *
1061 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001062 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001063static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1064 struct nameidata *nd, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001065{
Nick Pigginbaa03892010-08-18 04:37:31 +10001066 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001067 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001068
Miklos Szeredibad61182012-03-26 12:54:24 +02001069 *need_lookup = false;
1070 dentry = d_lookup(dir, name);
1071 if (dentry) {
1072 if (d_need_lookup(dentry)) {
1073 *need_lookup = true;
1074 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1075 error = d_revalidate(dentry, nd);
1076 if (unlikely(error <= 0)) {
1077 if (error < 0) {
1078 dput(dentry);
1079 return ERR_PTR(error);
1080 } else if (!d_invalidate(dentry)) {
1081 dput(dentry);
1082 dentry = NULL;
1083 }
1084 }
1085 }
1086 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001087
Miklos Szeredibad61182012-03-26 12:54:24 +02001088 if (!dentry) {
1089 dentry = d_alloc(dir, name);
1090 if (unlikely(!dentry))
1091 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001092
Miklos Szeredibad61182012-03-26 12:54:24 +02001093 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001094 }
1095 return dentry;
1096}
1097
1098/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001099 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1100 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1101 *
1102 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001103 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001104static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1105 struct nameidata *nd)
Josef Bacik44396f42011-05-31 11:58:49 -04001106{
Josef Bacik44396f42011-05-31 11:58:49 -04001107 struct dentry *old;
1108
1109 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001110 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001111 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001112 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001113 }
Josef Bacik44396f42011-05-31 11:58:49 -04001114
Miklos Szeredibad61182012-03-26 12:54:24 +02001115 old = dir->i_op->lookup(dir, dentry, nd);
Josef Bacik44396f42011-05-31 11:58:49 -04001116 if (unlikely(old)) {
1117 dput(dentry);
1118 dentry = old;
1119 }
1120 return dentry;
1121}
1122
Al Viroa3255542012-03-30 14:41:51 -04001123static struct dentry *__lookup_hash(struct qstr *name,
1124 struct dentry *base, struct nameidata *nd)
1125{
Miklos Szeredibad61182012-03-26 12:54:24 +02001126 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001127 struct dentry *dentry;
1128
Miklos Szeredibad61182012-03-26 12:54:24 +02001129 dentry = lookup_dcache(name, base, nd, &need_lookup);
1130 if (!need_lookup)
1131 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001132
Miklos Szeredibad61182012-03-26 12:54:24 +02001133 return lookup_real(base->d_inode, dentry, nd);
Al Viroa3255542012-03-30 14:41:51 -04001134}
1135
Josef Bacik44396f42011-05-31 11:58:49 -04001136/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 * It's more convoluted than I'd like it to be, but... it's still fairly
1138 * small and for now I'd prefer to have fast path as straight as possible.
1139 * It _is_ time-critical.
1140 */
1141static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001142 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
Jan Blunck4ac91372008-02-14 19:34:32 -08001144 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001145 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001146 int need_reval = 1;
1147 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001148 int err;
1149
Al Viro3cac2602009-08-13 18:27:43 +04001150 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001151 * Rename seqlock is not required here because in the off chance
1152 * of a false negative due to a concurrent rename, we're going to
1153 * do the non-racy lookup, below.
1154 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001155 if (nd->flags & LOOKUP_RCU) {
1156 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001157 *inode = nd->inode;
1158 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001159 if (!dentry)
1160 goto unlazy;
1161
Nick Piggin31e6b012011-01-07 17:49:52 +11001162 /* Memory barrier in read_seqcount_begin of child is enough */
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001165 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001166
Miklos Szeredifa4ee152012-03-26 12:54:19 +02001167 if (unlikely(d_need_lookup(dentry)))
1168 goto unlazy;
Al Viro24643082011-02-15 01:26:22 -05001169 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001170 status = d_revalidate(dentry, nd);
1171 if (unlikely(status <= 0)) {
1172 if (status != -ECHILD)
1173 need_reval = 0;
1174 goto unlazy;
1175 }
Al Viro24643082011-02-15 01:26:22 -05001176 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001177 path->mnt = mnt;
1178 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001179 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1180 goto unlazy;
1181 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1182 goto unlazy;
1183 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001184unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001185 if (unlazy_walk(nd, dentry))
1186 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001187 } else {
1188 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001189 }
Al Viro5a18fff2011-03-11 04:44:53 -05001190
Al Viro81e6f522012-03-30 14:48:04 -04001191 if (unlikely(!dentry))
1192 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001193
Al Viro81e6f522012-03-30 14:48:04 -04001194 if (unlikely(d_need_lookup(dentry))) {
1195 dput(dentry);
1196 goto need_lookup;
Al Viro24643082011-02-15 01:26:22 -05001197 }
Al Viro81e6f522012-03-30 14:48:04 -04001198
Al Viro5a18fff2011-03-11 04:44:53 -05001199 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1200 status = d_revalidate(dentry, nd);
1201 if (unlikely(status <= 0)) {
1202 if (status < 0) {
1203 dput(dentry);
1204 return status;
1205 }
1206 if (!d_invalidate(dentry)) {
1207 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001208 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001209 }
1210 }
Al Viro3f6c7c72012-03-30 14:04:16 -04001211done:
David Howells9875cf82011-01-14 18:45:21 +00001212 path->mnt = mnt;
1213 path->dentry = dentry;
1214 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001215 if (unlikely(err < 0)) {
1216 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001217 return err;
Ian Kent89312212011-01-18 12:06:10 +08001218 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001219 if (err)
1220 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001221 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001223
1224need_lookup:
1225 BUG_ON(nd->inode != parent->d_inode);
1226
1227 mutex_lock(&parent->d_inode->i_mutex);
1228 dentry = __lookup_hash(name, parent, nd);
1229 mutex_unlock(&parent->d_inode->i_mutex);
1230 if (IS_ERR(dentry))
1231 return PTR_ERR(dentry);
1232 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233}
1234
Al Viro52094c82011-02-21 21:34:47 -05001235static inline int may_lookup(struct nameidata *nd)
1236{
1237 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001238 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001239 if (err != -ECHILD)
1240 return err;
Al Viro19660af2011-03-25 10:32:48 -04001241 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001242 return -ECHILD;
1243 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001244 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001245}
1246
Al Viro9856fa12011-03-04 14:22:06 -05001247static inline int handle_dots(struct nameidata *nd, int type)
1248{
1249 if (type == LAST_DOTDOT) {
1250 if (nd->flags & LOOKUP_RCU) {
1251 if (follow_dotdot_rcu(nd))
1252 return -ECHILD;
1253 } else
1254 follow_dotdot(nd);
1255 }
1256 return 0;
1257}
1258
Al Viro951361f2011-03-04 14:44:37 -05001259static void terminate_walk(struct nameidata *nd)
1260{
1261 if (!(nd->flags & LOOKUP_RCU)) {
1262 path_put(&nd->path);
1263 } else {
1264 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001265 if (!(nd->flags & LOOKUP_ROOT))
1266 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001267 rcu_read_unlock();
1268 br_read_unlock(vfsmount_lock);
1269 }
1270}
1271
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001272/*
1273 * Do we need to follow links? We _really_ want to be able
1274 * to do this check without having to look at inode->i_op,
1275 * so we keep a cache of "no, this doesn't need follow_link"
1276 * for the common case.
1277 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001278static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001279{
1280 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1281 if (likely(inode->i_op->follow_link))
1282 return follow;
1283
1284 /* This gets set once for the inode lifetime */
1285 spin_lock(&inode->i_lock);
1286 inode->i_opflags |= IOP_NOFOLLOW;
1287 spin_unlock(&inode->i_lock);
1288 }
1289 return 0;
1290}
1291
Al Viroce57dfc2011-03-13 19:58:58 -04001292static inline int walk_component(struct nameidata *nd, struct path *path,
1293 struct qstr *name, int type, int follow)
1294{
1295 struct inode *inode;
1296 int err;
1297 /*
1298 * "." and ".." are special - ".." especially so because it has
1299 * to be able to know about the current root directory and
1300 * parent relationships.
1301 */
1302 if (unlikely(type != LAST_NORM))
1303 return handle_dots(nd, type);
1304 err = do_lookup(nd, name, path, &inode);
1305 if (unlikely(err)) {
1306 terminate_walk(nd);
1307 return err;
1308 }
1309 if (!inode) {
1310 path_to_nameidata(path, nd);
1311 terminate_walk(nd);
1312 return -ENOENT;
1313 }
Linus Torvalds7813b942011-08-07 09:53:20 -07001314 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001315 if (nd->flags & LOOKUP_RCU) {
1316 if (unlikely(unlazy_walk(nd, path->dentry))) {
1317 terminate_walk(nd);
1318 return -ECHILD;
1319 }
1320 }
Al Viroce57dfc2011-03-13 19:58:58 -04001321 BUG_ON(inode != path->dentry->d_inode);
1322 return 1;
1323 }
1324 path_to_nameidata(path, nd);
1325 nd->inode = inode;
1326 return 0;
1327}
1328
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329/*
Al Virob3563792011-03-14 21:54:55 -04001330 * This limits recursive symlink follows to 8, while
1331 * limiting consecutive symlinks to 40.
1332 *
1333 * Without that kind of total limit, nasty chains of consecutive
1334 * symlinks can cause almost arbitrarily long lookups.
1335 */
1336static inline int nested_symlink(struct path *path, struct nameidata *nd)
1337{
1338 int res;
1339
Al Virob3563792011-03-14 21:54:55 -04001340 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1341 path_put_conditional(path, nd);
1342 path_put(&nd->path);
1343 return -ELOOP;
1344 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001345 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001346
1347 nd->depth++;
1348 current->link_count++;
1349
1350 do {
1351 struct path link = *path;
1352 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001353
1354 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001355 if (!res)
1356 res = walk_component(nd, path, &nd->last,
1357 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001358 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001359 } while (res > 0);
1360
1361 current->link_count--;
1362 nd->depth--;
1363 return res;
1364}
1365
1366/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001367 * We really don't want to look at inode->i_op->lookup
1368 * when we don't have to. So we keep a cache bit in
1369 * the inode ->i_opflags field that says "yes, we can
1370 * do lookup on this inode".
1371 */
1372static inline int can_lookup(struct inode *inode)
1373{
1374 if (likely(inode->i_opflags & IOP_LOOKUP))
1375 return 1;
1376 if (likely(!inode->i_op->lookup))
1377 return 0;
1378
1379 /* We do this once for the lifetime of the inode */
1380 spin_lock(&inode->i_lock);
1381 inode->i_opflags |= IOP_LOOKUP;
1382 spin_unlock(&inode->i_lock);
1383 return 1;
1384}
1385
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001386/*
1387 * We can do the critical dentry name comparison and hashing
1388 * operations one word at a time, but we are limited to:
1389 *
1390 * - Architectures with fast unaligned word accesses. We could
1391 * do a "get_unaligned()" if this helps and is sufficiently
1392 * fast.
1393 *
1394 * - Little-endian machines (so that we can generate the mask
1395 * of low bytes efficiently). Again, we *could* do a byte
1396 * swapping load on big-endian architectures if that is not
1397 * expensive enough to make the optimization worthless.
1398 *
1399 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1400 * do not trap on the (extremely unlikely) case of a page
1401 * crossing operation.
1402 *
1403 * - Furthermore, we need an efficient 64-bit compile for the
1404 * 64-bit case in order to generate the "number of bytes in
1405 * the final mask". Again, that could be replaced with a
1406 * efficient population count instruction or similar.
1407 */
1408#ifdef CONFIG_DCACHE_WORD_ACCESS
1409
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001410#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001411
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001412#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001413
1414static inline unsigned int fold_hash(unsigned long hash)
1415{
1416 hash += hash >> (8*sizeof(int));
1417 return hash;
1418}
1419
1420#else /* 32-bit case */
1421
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001422#define fold_hash(x) (x)
1423
1424#endif
1425
1426unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1427{
1428 unsigned long a, mask;
1429 unsigned long hash = 0;
1430
1431 for (;;) {
1432 a = *(unsigned long *)name;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001433 if (len < sizeof(unsigned long))
1434 break;
1435 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001436 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001437 name += sizeof(unsigned long);
1438 len -= sizeof(unsigned long);
1439 if (!len)
1440 goto done;
1441 }
1442 mask = ~(~0ul << len*8);
1443 hash += mask & a;
1444done:
1445 return fold_hash(hash);
1446}
1447EXPORT_SYMBOL(full_name_hash);
1448
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001449/*
1450 * Calculate the length and hash of the path component, and
1451 * return the length of the component;
1452 */
1453static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1454{
1455 unsigned long a, mask, hash, len;
1456
1457 hash = a = 0;
1458 len = -sizeof(unsigned long);
1459 do {
1460 hash = (hash + a) * 9;
1461 len += sizeof(unsigned long);
1462 a = *(unsigned long *)(name+len);
1463 /* Do we have any NUL or '/' bytes in this word? */
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001464 mask = has_zero(a) | has_zero(a ^ REPEAT_BYTE('/'));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001465 } while (!mask);
1466
1467 /* The mask *below* the first high bit set */
1468 mask = (mask - 1) & ~mask;
1469 mask >>= 7;
1470 hash += a & mask;
1471 *hashp = fold_hash(hash);
1472
1473 return len + count_masked_bytes(mask);
1474}
1475
1476#else
1477
Linus Torvalds0145acc2012-03-02 14:32:59 -08001478unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1479{
1480 unsigned long hash = init_name_hash();
1481 while (len--)
1482 hash = partial_name_hash(*name++, hash);
1483 return end_name_hash(hash);
1484}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001485EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001486
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001487/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001488 * We know there's a real path component here of at least
1489 * one character.
1490 */
1491static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1492{
1493 unsigned long hash = init_name_hash();
1494 unsigned long len = 0, c;
1495
1496 c = (unsigned char)*name;
1497 do {
1498 len++;
1499 hash = partial_name_hash(c, hash);
1500 c = (unsigned char)name[len];
1501 } while (c && c != '/');
1502 *hashp = end_name_hash(hash);
1503 return len;
1504}
1505
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001506#endif
1507
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001508/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001510 * This is the basic name resolution function, turning a pathname into
1511 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001513 * Returns 0 and nd will have valid dentry and mnt on success.
1514 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 */
Al Viro6de88d72009-08-09 01:41:57 +04001516static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
1518 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521 while (*name=='/')
1522 name++;
1523 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001524 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 /* At this point we know we have a real path component. */
1527 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001529 long len;
Al Virofe479a52011-02-22 15:10:03 -05001530 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Al Viro52094c82011-02-21 21:34:47 -05001532 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 if (err)
1534 break;
1535
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001536 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001538 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
Al Virofe479a52011-02-22 15:10:03 -05001540 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001541 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001542 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001543 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001544 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001545 nd->flags |= LOOKUP_JUMPED;
1546 }
Al Virofe479a52011-02-22 15:10:03 -05001547 break;
1548 case 1:
1549 type = LAST_DOT;
1550 }
Al Viro5a202bc2011-03-08 14:17:44 -05001551 if (likely(type == LAST_NORM)) {
1552 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001553 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001554 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1555 err = parent->d_op->d_hash(parent, nd->inode,
1556 &this);
1557 if (err < 0)
1558 break;
1559 }
1560 }
Al Virofe479a52011-02-22 15:10:03 -05001561
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001562 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001564 /*
1565 * If it wasn't NUL, we know it was '/'. Skip that
1566 * slash, and continue until no more slashes.
1567 */
1568 do {
1569 len++;
1570 } while (unlikely(name[len] == '/'));
1571 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001572 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001573 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Al Viroce57dfc2011-03-13 19:58:58 -04001575 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1576 if (err < 0)
1577 return err;
Al Virofe479a52011-02-22 15:10:03 -05001578
Al Viroce57dfc2011-03-13 19:58:58 -04001579 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001580 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001582 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001583 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001584 if (can_lookup(nd->inode))
1585 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001587 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 /* here ends the main loop */
1589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590last_component:
Al Virob3563792011-03-14 21:54:55 -04001591 nd->last = this;
1592 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001593 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 }
Al Viro951361f2011-03-04 14:44:37 -05001595 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 return err;
1597}
1598
Al Viro70e9b352011-03-05 21:12:22 -05001599static int path_init(int dfd, const char *name, unsigned int flags,
1600 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001602 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001603 int fput_needed;
1604 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001607 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001609 if (flags & LOOKUP_ROOT) {
1610 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001611 if (*name) {
1612 if (!inode->i_op->lookup)
1613 return -ENOTDIR;
1614 retval = inode_permission(inode, MAY_EXEC);
1615 if (retval)
1616 return retval;
1617 }
Al Viro5b6ca022011-03-09 23:04:47 -05001618 nd->path = nd->root;
1619 nd->inode = inode;
1620 if (flags & LOOKUP_RCU) {
1621 br_read_lock(vfsmount_lock);
1622 rcu_read_lock();
1623 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1624 } else {
1625 path_get(&nd->path);
1626 }
1627 return 0;
1628 }
1629
Al Viro2a737872009-04-07 11:49:53 -04001630 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001633 if (flags & LOOKUP_RCU) {
1634 br_read_lock(vfsmount_lock);
1635 rcu_read_lock();
1636 set_root_rcu(nd);
1637 } else {
1638 set_root(nd);
1639 path_get(&nd->root);
1640 }
Al Viro2a737872009-04-07 11:49:53 -04001641 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001642 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001643 if (flags & LOOKUP_RCU) {
1644 struct fs_struct *fs = current->fs;
1645 unsigned seq;
1646
1647 br_read_lock(vfsmount_lock);
1648 rcu_read_lock();
1649
1650 do {
1651 seq = read_seqcount_begin(&fs->seq);
1652 nd->path = fs->pwd;
1653 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1654 } while (read_seqcount_retry(&fs->seq, seq));
1655 } else {
1656 get_fs_pwd(current->fs, &nd->path);
1657 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001658 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001659 struct dentry *dentry;
1660
Al Viro1abf0c72011-03-13 03:51:11 -04001661 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001662 retval = -EBADF;
1663 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001664 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001665
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001666 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001667
Al Virof52e0c12011-03-14 18:56:51 -04001668 if (*name) {
1669 retval = -ENOTDIR;
1670 if (!S_ISDIR(dentry->d_inode->i_mode))
1671 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001672
Al Viro4ad5abb2011-06-20 19:57:03 -04001673 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001674 if (retval)
1675 goto fput_fail;
1676 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001677
Jan Blunck5dd784d02008-02-14 19:34:38 -08001678 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001679 if (flags & LOOKUP_RCU) {
1680 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001681 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001682 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1683 br_read_lock(vfsmount_lock);
1684 rcu_read_lock();
1685 } else {
1686 path_get(&file->f_path);
1687 fput_light(file, fput_needed);
1688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 }
Al Viroe41f7d42011-02-22 14:02:58 -05001690
Nick Piggin31e6b012011-01-07 17:49:52 +11001691 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001692 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001693
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001694fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001695 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001696out_fail:
1697 return retval;
1698}
1699
Al Virobd92d7f2011-03-14 19:54:59 -04001700static inline int lookup_last(struct nameidata *nd, struct path *path)
1701{
1702 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1703 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1704
1705 nd->flags &= ~LOOKUP_PARENT;
1706 return walk_component(nd, path, &nd->last, nd->last_type,
1707 nd->flags & LOOKUP_FOLLOW);
1708}
1709
Al Viro9b4a9b12009-04-07 11:44:16 -04001710/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001711static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001712 unsigned int flags, struct nameidata *nd)
1713{
Al Viro70e9b352011-03-05 21:12:22 -05001714 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001715 struct path path;
1716 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001717
1718 /*
1719 * Path walking is largely split up into 2 different synchronisation
1720 * schemes, rcu-walk and ref-walk (explained in
1721 * Documentation/filesystems/path-lookup.txt). These share much of the
1722 * path walk code, but some things particularly setup, cleanup, and
1723 * following mounts are sufficiently divergent that functions are
1724 * duplicated. Typically there is a function foo(), and its RCU
1725 * analogue, foo_rcu().
1726 *
1727 * -ECHILD is the error number of choice (just to avoid clashes) that
1728 * is returned if some aspect of an rcu-walk fails. Such an error must
1729 * be handled by restarting a traditional ref-walk (which will always
1730 * be able to complete).
1731 */
Al Virobd92d7f2011-03-14 19:54:59 -04001732 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001733
Al Virobd92d7f2011-03-14 19:54:59 -04001734 if (unlikely(err))
1735 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001736
1737 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001738 err = link_path_walk(name, nd);
1739
1740 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001741 err = lookup_last(nd, &path);
1742 while (err > 0) {
1743 void *cookie;
1744 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001745 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001746 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001747 if (!err)
1748 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001749 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001750 }
1751 }
Al Viroee0827c2011-02-21 23:38:09 -05001752
Al Viro9f1fafe2011-03-25 11:00:12 -04001753 if (!err)
1754 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001755
1756 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1757 if (!nd->inode->i_op->lookup) {
1758 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001759 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001760 }
1761 }
Al Viro16c2cd72011-02-22 15:50:10 -05001762
Al Viro70e9b352011-03-05 21:12:22 -05001763 if (base)
1764 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001765
Al Viro5b6ca022011-03-09 23:04:47 -05001766 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001767 path_put(&nd->root);
1768 nd->root.mnt = NULL;
1769 }
Al Virobd92d7f2011-03-14 19:54:59 -04001770 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001771}
Nick Piggin31e6b012011-01-07 17:49:52 +11001772
Al Viroee0827c2011-02-21 23:38:09 -05001773static int do_path_lookup(int dfd, const char *name,
1774 unsigned int flags, struct nameidata *nd)
1775{
1776 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1777 if (unlikely(retval == -ECHILD))
1778 retval = path_lookupat(dfd, name, flags, nd);
1779 if (unlikely(retval == -ESTALE))
1780 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001781
1782 if (likely(!retval)) {
1783 if (unlikely(!audit_dummy_context())) {
1784 if (nd->path.dentry && nd->inode)
1785 audit_inode(name, nd->path.dentry);
1786 }
1787 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001788 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789}
1790
Al Viroc9c6cac2011-02-16 15:15:47 -05001791int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001792{
Al Viroc9c6cac2011-02-16 15:15:47 -05001793 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001794}
1795
Al Virod1811462008-08-02 00:49:18 -04001796int kern_path(const char *name, unsigned int flags, struct path *path)
1797{
1798 struct nameidata nd;
1799 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1800 if (!res)
1801 *path = nd.path;
1802 return res;
1803}
1804
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001805/**
1806 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1807 * @dentry: pointer to dentry of the base directory
1808 * @mnt: pointer to vfs mount of the base directory
1809 * @name: pointer to file name
1810 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001811 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001812 */
1813int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1814 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001815 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001816{
Al Viroe0a01242011-06-27 17:00:37 -04001817 struct nameidata nd;
1818 int err;
1819 nd.root.dentry = dentry;
1820 nd.root.mnt = mnt;
1821 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001822 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001823 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1824 if (!err)
1825 *path = nd.path;
1826 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001827}
1828
James Morris057f6c02007-04-26 00:12:05 -07001829/*
1830 * Restricted form of lookup. Doesn't follow links, single-component only,
1831 * needs parent already locked. Doesn't follow mounts.
1832 * SMP-safe.
1833 */
Adrian Bunka244e162006-03-31 02:32:11 -08001834static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835{
Jan Blunck4ac91372008-02-14 19:34:32 -08001836 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837}
1838
Christoph Hellwigeead1912007-10-16 23:25:38 -07001839/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001840 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001841 * @name: pathname component to lookup
1842 * @base: base directory to lookup from
1843 * @len: maximum length @len should be interpreted to
1844 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001845 * Note that this routine is purely a helper for filesystem usage and should
1846 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001847 * nameidata argument is passed to the filesystem methods and a filesystem
1848 * using this helper needs to be prepared for that.
1849 */
James Morris057f6c02007-04-26 00:12:05 -07001850struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1851{
James Morris057f6c02007-04-26 00:12:05 -07001852 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001853 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02001854 int err;
James Morris057f6c02007-04-26 00:12:05 -07001855
David Woodhouse2f9092e2009-04-20 23:18:37 +01001856 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1857
Al Viro6a96ba52011-03-07 23:49:20 -05001858 this.name = name;
1859 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08001860 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05001861 if (!len)
1862 return ERR_PTR(-EACCES);
1863
Al Viro6a96ba52011-03-07 23:49:20 -05001864 while (len--) {
1865 c = *(const unsigned char *)name++;
1866 if (c == '/' || c == '\0')
1867 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05001868 }
Al Viro5a202bc2011-03-08 14:17:44 -05001869 /*
1870 * See if the low-level filesystem might want
1871 * to use its own hash..
1872 */
1873 if (base->d_flags & DCACHE_OP_HASH) {
1874 int err = base->d_op->d_hash(base, base->d_inode, &this);
1875 if (err < 0)
1876 return ERR_PTR(err);
1877 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001878
Miklos Szeredicda309d2012-03-26 12:54:21 +02001879 err = inode_permission(base->d_inode, MAY_EXEC);
1880 if (err)
1881 return ERR_PTR(err);
1882
Christoph Hellwig49705b72005-11-08 21:35:06 -08001883 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001884}
1885
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001886int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1887 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888{
Al Viro2d8f3032008-07-22 09:59:21 -04001889 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001890 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001893
1894 BUG_ON(flags & LOOKUP_PARENT);
1895
1896 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001898 if (!err)
1899 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 }
1901 return err;
1902}
1903
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001904int user_path_at(int dfd, const char __user *name, unsigned flags,
1905 struct path *path)
1906{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07001907 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001908}
1909
Al Viro2ad94ae2008-07-21 09:32:51 -04001910static int user_path_parent(int dfd, const char __user *path,
1911 struct nameidata *nd, char **name)
1912{
1913 char *s = getname(path);
1914 int error;
1915
1916 if (IS_ERR(s))
1917 return PTR_ERR(s);
1918
1919 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1920 if (error)
1921 putname(s);
1922 else
1923 *name = s;
1924
1925 return error;
1926}
1927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928/*
1929 * It's inline, so penalty for filesystems that don't use sticky bit is
1930 * minimal.
1931 */
1932static inline int check_sticky(struct inode *dir, struct inode *inode)
1933{
David Howellsda9592e2008-11-14 10:39:05 +11001934 uid_t fsuid = current_fsuid();
1935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 if (!(dir->i_mode & S_ISVTX))
1937 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001938 if (current_user_ns() != inode_userns(inode))
1939 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001940 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001942 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001944
1945other_userns:
1946 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947}
1948
1949/*
1950 * Check whether we can remove a link victim from directory dir, check
1951 * whether the type of victim is right.
1952 * 1. We can't do it if dir is read-only (done in permission())
1953 * 2. We should have write and exec permissions on dir
1954 * 3. We can't remove anything from append-only dir
1955 * 4. We can't do anything with immutable dir (done in permission())
1956 * 5. If the sticky bit on dir is set we should either
1957 * a. be owner of dir, or
1958 * b. be owner of victim, or
1959 * c. have CAP_FOWNER capability
1960 * 6. If the victim is append-only or immutable we can't do antyhing with
1961 * links pointing to it.
1962 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1963 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1964 * 9. We can't remove a root or mountpoint.
1965 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1966 * nfs_async_unlink().
1967 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001968static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969{
1970 int error;
1971
1972 if (!victim->d_inode)
1973 return -ENOENT;
1974
1975 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001976 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
Al Virof419a2e2008-07-22 00:07:17 -04001978 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 if (error)
1980 return error;
1981 if (IS_APPEND(dir))
1982 return -EPERM;
1983 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001984 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 return -EPERM;
1986 if (isdir) {
1987 if (!S_ISDIR(victim->d_inode->i_mode))
1988 return -ENOTDIR;
1989 if (IS_ROOT(victim))
1990 return -EBUSY;
1991 } else if (S_ISDIR(victim->d_inode->i_mode))
1992 return -EISDIR;
1993 if (IS_DEADDIR(dir))
1994 return -ENOENT;
1995 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1996 return -EBUSY;
1997 return 0;
1998}
1999
2000/* Check whether we can create an object with dentry child in directory
2001 * dir.
2002 * 1. We can't do it if child already exists (open has special treatment for
2003 * this case, but since we are inlined it's OK)
2004 * 2. We can't do it if dir is read-only (done in permission())
2005 * 3. We should have write and exec permissions on dir
2006 * 4. We can't do it if dir is immutable (done in permission())
2007 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002008static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
2010 if (child->d_inode)
2011 return -EEXIST;
2012 if (IS_DEADDIR(dir))
2013 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002014 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015}
2016
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017/*
2018 * p1 and p2 should be directories on the same fs.
2019 */
2020struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2021{
2022 struct dentry *p;
2023
2024 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002025 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 return NULL;
2027 }
2028
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002029 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002031 p = d_ancestor(p2, p1);
2032 if (p) {
2033 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2034 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2035 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 }
2037
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002038 p = d_ancestor(p1, p2);
2039 if (p) {
2040 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2041 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2042 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
2044
Ingo Molnarf2eace22006-07-03 00:25:05 -07002045 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2046 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return NULL;
2048}
2049
2050void unlock_rename(struct dentry *p1, struct dentry *p2)
2051{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002052 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002054 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002055 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 }
2057}
2058
Al Viro4acdaf22011-07-26 01:42:34 -04002059int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 struct nameidata *nd)
2061{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002062 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
2064 if (error)
2065 return error;
2066
Al Viroacfa4382008-12-04 10:06:33 -05002067 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 return -EACCES; /* shouldn't it be ENOSYS? */
2069 mode &= S_IALLUGO;
2070 mode |= S_IFREG;
2071 error = security_inode_create(dir, dentry, mode);
2072 if (error)
2073 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002075 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002076 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 return error;
2078}
2079
Al Viro73d049a2011-03-11 12:08:24 -05002080static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002082 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 struct inode *inode = dentry->d_inode;
2084 int error;
2085
Al Virobcda7652011-03-13 16:42:14 -04002086 /* O_PATH? */
2087 if (!acc_mode)
2088 return 0;
2089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 if (!inode)
2091 return -ENOENT;
2092
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002093 switch (inode->i_mode & S_IFMT) {
2094 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002096 case S_IFDIR:
2097 if (acc_mode & MAY_WRITE)
2098 return -EISDIR;
2099 break;
2100 case S_IFBLK:
2101 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002102 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002104 /*FALLTHRU*/
2105 case S_IFIFO:
2106 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002108 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002109 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002110
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002111 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002112 if (error)
2113 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 /*
2116 * An append-only file must be opened in append mode for writing.
2117 */
2118 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002119 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002120 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002122 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 }
2124
2125 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002126 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002127 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002129 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002130}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Jeff Laytone1181ee2010-12-07 16:19:50 -05002132static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002133{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002134 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002135 struct inode *inode = path->dentry->d_inode;
2136 int error = get_write_access(inode);
2137 if (error)
2138 return error;
2139 /*
2140 * Refuse to truncate files with mandatory locks held on them.
2141 */
2142 error = locks_verify_locked(inode);
2143 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002144 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002145 if (!error) {
2146 error = do_truncate(path->dentry, 0,
2147 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002148 filp);
Al Viro7715b522009-12-16 03:54:00 -05002149 }
2150 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002151 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152}
2153
Dave Hansend57999e2008-02-15 14:37:27 -08002154static inline int open_to_namei_flags(int flag)
2155{
Al Viro8a5e9292011-06-25 19:15:54 -04002156 if ((flag & O_ACCMODE) == 3)
2157 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002158 return flag;
2159}
2160
Nick Piggin31e6b012011-01-07 17:49:52 +11002161/*
Al Virofe2d35f2011-03-05 22:58:25 -05002162 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002163 */
Al Virofb1cc552009-12-24 01:58:28 -05002164static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002165 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002166{
Al Viroa1e28032009-12-24 02:12:06 -05002167 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002168 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002169 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002170 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002171 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002172 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002173 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002174 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002175
Al Viroc3e380b2011-02-23 13:39:45 -05002176 nd->flags &= ~LOOKUP_PARENT;
2177 nd->flags |= op->intent;
2178
Al Viro1f36f772009-12-26 10:56:19 -05002179 switch (nd->last_type) {
2180 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002181 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002182 error = handle_dots(nd, nd->last_type);
2183 if (error)
2184 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002185 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002186 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002187 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002188 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002189 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002190 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002191 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002192 error = -EISDIR;
2193 goto exit;
2194 }
2195 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002196 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002197 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002198 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002199 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002200 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002201 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002202 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002203
Al Viroca344a892011-03-09 00:36:45 -05002204 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002205 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002206 if (nd->last.name[nd->last.len])
2207 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002208 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2209 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002210 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002211 error = walk_component(nd, path, &nd->last, LAST_NORM,
2212 !symlink_ok);
2213 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002214 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002215 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002216 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002217 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002218 error = complete_walk(nd);
2219 if (error)
Miklos Szeredi7f6c7e62012-03-06 13:56:34 +01002220 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002221
2222 error = -ENOTDIR;
2223 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002224 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002225 goto exit;
2226 }
2227 audit_inode(pathname, nd->path.dentry);
2228 goto ok;
2229 }
2230
2231 /* create side of things */
Al Viroa3fbbde2011-11-07 21:21:26 +00002232 /*
2233 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2234 * cleared when we got to the last component we are about to look up
2235 */
Al Viro9f1fafe2011-03-25 11:00:12 -04002236 error = complete_walk(nd);
2237 if (error)
2238 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002239
2240 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002241 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002242 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002243 if (nd->last.name[nd->last.len])
2244 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002245
Al Viroa1e28032009-12-24 02:12:06 -05002246 mutex_lock(&dir->d_inode->i_mutex);
2247
Al Viro6c0d46c2011-03-09 00:59:59 -05002248 dentry = lookup_hash(nd);
2249 error = PTR_ERR(dentry);
2250 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002251 mutex_unlock(&dir->d_inode->i_mutex);
2252 goto exit;
2253 }
2254
Al Viro6c0d46c2011-03-09 00:59:59 -05002255 path->dentry = dentry;
2256 path->mnt = nd->path.mnt;
2257
Al Virofb1cc552009-12-24 01:58:28 -05002258 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002259 if (!dentry->d_inode) {
Al Viroa218d0f2011-11-21 14:59:34 -05002260 umode_t mode = op->mode;
Al Viro6c0d46c2011-03-09 00:59:59 -05002261 if (!IS_POSIXACL(dir->d_inode))
2262 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002263 /*
2264 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002265 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002266 * the time when the file is created and when
2267 * a permanent write count is taken through
2268 * the 'struct file' in nameidata_to_filp().
2269 */
2270 error = mnt_want_write(nd->path.mnt);
2271 if (error)
2272 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002273 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002274 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002275 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002276 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002277 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002278 error = security_path_mknod(&nd->path, dentry, mode, 0);
2279 if (error)
2280 goto exit_mutex_unlock;
2281 error = vfs_create(dir->d_inode, dentry, mode, nd);
2282 if (error)
2283 goto exit_mutex_unlock;
2284 mutex_unlock(&dir->d_inode->i_mutex);
2285 dput(nd->path.dentry);
2286 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002287 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002288 }
2289
2290 /*
2291 * It already exists.
2292 */
2293 mutex_unlock(&dir->d_inode->i_mutex);
2294 audit_inode(pathname, path->dentry);
2295
2296 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002297 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002298 goto exit_dput;
2299
David Howells9875cf82011-01-14 18:45:21 +00002300 error = follow_managed(path, nd->flags);
2301 if (error < 0)
2302 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002303
Al Viroa3fbbde2011-11-07 21:21:26 +00002304 if (error)
2305 nd->flags |= LOOKUP_JUMPED;
2306
Al Virofb1cc552009-12-24 01:58:28 -05002307 error = -ENOENT;
2308 if (!path->dentry->d_inode)
2309 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002310
2311 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002312 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002313
2314 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002315 nd->inode = path->dentry->d_inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002316 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2317 error = complete_walk(nd);
2318 if (error)
Miklos Szeredi097b1802012-03-06 13:56:33 +01002319 return ERR_PTR(error);
Al Virofb1cc552009-12-24 01:58:28 -05002320 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002321 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002322 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002323ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002324 if (!S_ISREG(nd->inode->i_mode))
2325 will_truncate = 0;
2326
Al Viro0f9d1a12011-03-09 00:13:14 -05002327 if (will_truncate) {
2328 error = mnt_want_write(nd->path.mnt);
2329 if (error)
2330 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002331 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002332 }
Al Viroca344a892011-03-09 00:36:45 -05002333common:
Al Virobcda7652011-03-13 16:42:14 -04002334 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002335 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002336 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002337 filp = nameidata_to_filp(nd);
2338 if (!IS_ERR(filp)) {
2339 error = ima_file_check(filp, op->acc_mode);
2340 if (error) {
2341 fput(filp);
2342 filp = ERR_PTR(error);
2343 }
2344 }
2345 if (!IS_ERR(filp)) {
2346 if (will_truncate) {
2347 error = handle_truncate(filp);
2348 if (error) {
2349 fput(filp);
2350 filp = ERR_PTR(error);
2351 }
2352 }
2353 }
Al Viroca344a892011-03-09 00:36:45 -05002354out:
2355 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002356 mnt_drop_write(nd->path.mnt);
2357 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002358 return filp;
2359
2360exit_mutex_unlock:
2361 mutex_unlock(&dir->d_inode->i_mutex);
2362exit_dput:
2363 path_put_conditional(path, nd);
2364exit:
Al Viroca344a892011-03-09 00:36:45 -05002365 filp = ERR_PTR(error);
2366 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002367}
2368
Al Viro13aab422011-02-23 17:54:08 -05002369static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002370 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371{
Al Virofe2d35f2011-03-05 22:58:25 -05002372 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002373 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002374 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002375 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002376
2377 filp = get_empty_filp();
2378 if (!filp)
2379 return ERR_PTR(-ENFILE);
2380
Al Viro47c805d2011-02-23 17:44:09 -05002381 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002382 nd->intent.open.file = filp;
2383 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2384 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002385
Al Viro73d049a2011-03-11 12:08:24 -05002386 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002387 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002388 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002389
Al Virofe2d35f2011-03-05 22:58:25 -05002390 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002391 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002392 if (unlikely(error))
2393 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394
Al Viro73d049a2011-03-11 12:08:24 -05002395 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002396 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002397 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002398 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002399 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002400 path_put_conditional(&path, nd);
2401 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002402 filp = ERR_PTR(-ELOOP);
2403 break;
2404 }
Al Viro73d049a2011-03-11 12:08:24 -05002405 nd->flags |= LOOKUP_PARENT;
2406 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002407 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002408 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002409 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002410 else
Al Viro73d049a2011-03-11 12:08:24 -05002411 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002412 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002413 }
Al Viro10fa8e62009-12-26 07:09:49 -05002414out:
Al Viro73d049a2011-03-11 12:08:24 -05002415 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2416 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002417 if (base)
2418 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002419 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002420 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Nick Piggin31e6b012011-01-07 17:49:52 +11002422out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002423 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002424 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425}
2426
Al Viro13aab422011-02-23 17:54:08 -05002427struct file *do_filp_open(int dfd, const char *pathname,
2428 const struct open_flags *op, int flags)
2429{
Al Viro73d049a2011-03-11 12:08:24 -05002430 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002431 struct file *filp;
2432
Al Viro73d049a2011-03-11 12:08:24 -05002433 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002434 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002435 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002436 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002437 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002438 return filp;
2439}
2440
Al Viro73d049a2011-03-11 12:08:24 -05002441struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2442 const char *name, const struct open_flags *op, int flags)
2443{
2444 struct nameidata nd;
2445 struct file *file;
2446
2447 nd.root.mnt = mnt;
2448 nd.root.dentry = dentry;
2449
2450 flags |= LOOKUP_ROOT;
2451
Al Virobcda7652011-03-13 16:42:14 -04002452 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002453 return ERR_PTR(-ELOOP);
2454
2455 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2456 if (unlikely(file == ERR_PTR(-ECHILD)))
2457 file = path_openat(-1, name, &nd, op, flags);
2458 if (unlikely(file == ERR_PTR(-ESTALE)))
2459 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2460 return file;
2461}
2462
Al Viroed75e952011-06-27 16:53:43 -04002463struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002465 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002466 struct nameidata nd;
2467 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2468 if (error)
2469 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002471 /*
2472 * Yucky last component or no last component at all?
2473 * (foo/., foo/.., /////)
2474 */
Al Viroed75e952011-06-27 16:53:43 -04002475 if (nd.last_type != LAST_NORM)
2476 goto out;
2477 nd.flags &= ~LOOKUP_PARENT;
2478 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2479 nd.intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002480
2481 /*
2482 * Do the final lookup.
2483 */
Al Viroed75e952011-06-27 16:53:43 -04002484 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2485 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 if (IS_ERR(dentry))
2487 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002488
Al Viroe9baf6e2008-05-15 04:49:12 -04002489 if (dentry->d_inode)
2490 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002491 /*
2492 * Special case - lookup gave negative, but... we had foo/bar/
2493 * From the vfs_mknod() POV we just have a negative dentry -
2494 * all is fine. Let's be bastards - you had / on the end, you've
2495 * been asking for (non-existent) directory. -ENOENT for you.
2496 */
Al Viroed75e952011-06-27 16:53:43 -04002497 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002498 dput(dentry);
2499 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002500 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002501 }
Al Viroed75e952011-06-27 16:53:43 -04002502 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002504eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002506 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507fail:
Al Viroed75e952011-06-27 16:53:43 -04002508 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2509out:
2510 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 return dentry;
2512}
Al Virodae6ad82011-06-26 11:50:15 -04002513EXPORT_SYMBOL(kern_path_create);
2514
2515struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2516{
2517 char *tmp = getname(pathname);
2518 struct dentry *res;
2519 if (IS_ERR(tmp))
2520 return ERR_CAST(tmp);
2521 res = kern_path_create(dfd, tmp, path, is_dir);
2522 putname(tmp);
2523 return res;
2524}
2525EXPORT_SYMBOL(user_path_create);
2526
Al Viro1a67aaf2011-07-26 01:52:52 -04002527int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002529 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
2531 if (error)
2532 return error;
2533
Serge E. Hallyne795b712011-03-23 16:43:25 -07002534 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2535 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 return -EPERM;
2537
Al Viroacfa4382008-12-04 10:06:33 -05002538 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 return -EPERM;
2540
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002541 error = devcgroup_inode_mknod(mode, dev);
2542 if (error)
2543 return error;
2544
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 error = security_inode_mknod(dir, dentry, mode, dev);
2546 if (error)
2547 return error;
2548
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002550 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002551 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return error;
2553}
2554
Al Virof69aac02011-07-26 04:31:40 -04002555static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08002556{
2557 switch (mode & S_IFMT) {
2558 case S_IFREG:
2559 case S_IFCHR:
2560 case S_IFBLK:
2561 case S_IFIFO:
2562 case S_IFSOCK:
2563 case 0: /* zero mode translates to S_IFREG */
2564 return 0;
2565 case S_IFDIR:
2566 return -EPERM;
2567 default:
2568 return -EINVAL;
2569 }
2570}
2571
Al Viro8208a222011-07-25 17:32:17 -04002572SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002573 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574{
Al Viro2ad94ae2008-07-21 09:32:51 -04002575 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002576 struct path path;
2577 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
2579 if (S_ISDIR(mode))
2580 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
Al Virodae6ad82011-06-26 11:50:15 -04002582 dentry = user_path_create(dfd, filename, &path, 0);
2583 if (IS_ERR(dentry))
2584 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002585
Al Virodae6ad82011-06-26 11:50:15 -04002586 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002587 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002588 error = may_mknod(mode);
2589 if (error)
2590 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002591 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002592 if (error)
2593 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002594 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002595 if (error)
2596 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002597 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 case 0: case S_IFREG:
Al Virodae6ad82011-06-26 11:50:15 -04002599 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 break;
2601 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002602 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 new_decode_dev(dev));
2604 break;
2605 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002606 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 }
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:
2612 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
2616 return error;
2617}
2618
Al Viro8208a222011-07-25 17:32:17 -04002619SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002620{
2621 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2622}
2623
Al Viro18bb1db2011-07-26 01:41:39 -04002624int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002626 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05002627 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
2629 if (error)
2630 return error;
2631
Al Viroacfa4382008-12-04 10:06:33 -05002632 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 return -EPERM;
2634
2635 mode &= (S_IRWXUGO|S_ISVTX);
2636 error = security_inode_mkdir(dir, dentry, mode);
2637 if (error)
2638 return error;
2639
Al Viro8de52772012-02-06 12:45:27 -05002640 if (max_links && dir->i_nlink >= max_links)
2641 return -EMLINK;
2642
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002644 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002645 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 return error;
2647}
2648
Al Viroa218d0f2011-11-21 14:59:34 -05002649SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650{
Dave Hansen6902d922006-09-30 23:29:01 -07002651 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002652 struct path path;
2653 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
Al Virodae6ad82011-06-26 11:50:15 -04002655 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07002656 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002657 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002658
Al Virodae6ad82011-06-26 11:50:15 -04002659 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002660 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04002661 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002662 if (error)
2663 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002664 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002665 if (error)
2666 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002667 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002668out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002669 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002670out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002671 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002672 mutex_unlock(&path.dentry->d_inode->i_mutex);
2673 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return error;
2675}
2676
Al Viroa218d0f2011-11-21 14:59:34 -05002677SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002678{
2679 return sys_mkdirat(AT_FDCWD, pathname, mode);
2680}
2681
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682/*
Sage Weila71905f2011-05-24 13:06:08 -07002683 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05002684 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07002685 * dentry, and if that is true (possibly after pruning the dcache),
2686 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 *
2688 * A low-level filesystem can, if it choses, legally
2689 * do a
2690 *
2691 * if (!d_unhashed(dentry))
2692 * return -EBUSY;
2693 *
2694 * if it cannot handle the case of removing a directory
2695 * that is still in use by something else..
2696 */
2697void dentry_unhash(struct dentry *dentry)
2698{
Vasily Averindc168422006-12-06 20:37:07 -08002699 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002701 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 __d_drop(dentry);
2703 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704}
2705
2706int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2707{
2708 int error = may_delete(dir, dentry, 1);
2709
2710 if (error)
2711 return error;
2712
Al Viroacfa4382008-12-04 10:06:33 -05002713 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 return -EPERM;
2715
Al Viro1d2ef592011-09-14 18:55:41 +01002716 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002717 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
Sage Weil912dbc12011-05-24 13:06:11 -07002719 error = -EBUSY;
2720 if (d_mountpoint(dentry))
2721 goto out;
2722
2723 error = security_inode_rmdir(dir, dentry);
2724 if (error)
2725 goto out;
2726
Sage Weil3cebde22011-05-29 21:20:59 -07002727 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002728 error = dir->i_op->rmdir(dir, dentry);
2729 if (error)
2730 goto out;
2731
2732 dentry->d_inode->i_flags |= S_DEAD;
2733 dont_mount(dentry);
2734
2735out:
2736 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01002737 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002738 if (!error)
2739 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 return error;
2741}
2742
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002743static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744{
2745 int error = 0;
2746 char * name;
2747 struct dentry *dentry;
2748 struct nameidata nd;
2749
Al Viro2ad94ae2008-07-21 09:32:51 -04002750 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002752 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
2754 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002755 case LAST_DOTDOT:
2756 error = -ENOTEMPTY;
2757 goto exit1;
2758 case LAST_DOT:
2759 error = -EINVAL;
2760 goto exit1;
2761 case LAST_ROOT:
2762 error = -EBUSY;
2763 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002765
2766 nd.flags &= ~LOOKUP_PARENT;
2767
Jan Blunck4ac91372008-02-14 19:34:32 -08002768 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002769 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002771 if (IS_ERR(dentry))
2772 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002773 if (!dentry->d_inode) {
2774 error = -ENOENT;
2775 goto exit3;
2776 }
Dave Hansen06227532008-02-15 14:37:34 -08002777 error = mnt_want_write(nd.path.mnt);
2778 if (error)
2779 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002780 error = security_path_rmdir(&nd.path, dentry);
2781 if (error)
2782 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002783 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002784exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002785 mnt_drop_write(nd.path.mnt);
2786exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002787 dput(dentry);
2788exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002789 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002791 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 putname(name);
2793 return error;
2794}
2795
Heiko Carstens3cdad422009-01-14 14:14:22 +01002796SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002797{
2798 return do_rmdir(AT_FDCWD, pathname);
2799}
2800
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801int vfs_unlink(struct inode *dir, struct dentry *dentry)
2802{
2803 int error = may_delete(dir, dentry, 0);
2804
2805 if (error)
2806 return error;
2807
Al Viroacfa4382008-12-04 10:06:33 -05002808 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 return -EPERM;
2810
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002811 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 if (d_mountpoint(dentry))
2813 error = -EBUSY;
2814 else {
2815 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002816 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002818 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002819 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002822 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823
2824 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2825 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002826 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 }
Robert Love0eeca282005-07-12 17:06:03 -04002829
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 return error;
2831}
2832
2833/*
2834 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002835 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 * writeout happening, and we don't want to prevent access to the directory
2837 * while waiting on the I/O.
2838 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002839static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840{
Al Viro2ad94ae2008-07-21 09:32:51 -04002841 int error;
2842 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 struct dentry *dentry;
2844 struct nameidata nd;
2845 struct inode *inode = NULL;
2846
Al Viro2ad94ae2008-07-21 09:32:51 -04002847 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002849 return error;
2850
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 error = -EISDIR;
2852 if (nd.last_type != LAST_NORM)
2853 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002854
2855 nd.flags &= ~LOOKUP_PARENT;
2856
Jan Blunck4ac91372008-02-14 19:34:32 -08002857 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002858 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 error = PTR_ERR(dentry);
2860 if (!IS_ERR(dentry)) {
2861 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002862 if (nd.last.name[nd.last.len])
2863 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002865 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002866 goto slashes;
2867 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002868 error = mnt_want_write(nd.path.mnt);
2869 if (error)
2870 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002871 error = security_path_unlink(&nd.path, dentry);
2872 if (error)
2873 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002874 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002875exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002876 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 exit2:
2878 dput(dentry);
2879 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002880 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 if (inode)
2882 iput(inode); /* truncate the inode here */
2883exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002884 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 putname(name);
2886 return error;
2887
2888slashes:
2889 error = !dentry->d_inode ? -ENOENT :
2890 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2891 goto exit2;
2892}
2893
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002894SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002895{
2896 if ((flag & ~AT_REMOVEDIR) != 0)
2897 return -EINVAL;
2898
2899 if (flag & AT_REMOVEDIR)
2900 return do_rmdir(dfd, pathname);
2901
2902 return do_unlinkat(dfd, pathname);
2903}
2904
Heiko Carstens3480b252009-01-14 14:14:16 +01002905SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002906{
2907 return do_unlinkat(AT_FDCWD, pathname);
2908}
2909
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002910int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002912 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913
2914 if (error)
2915 return error;
2916
Al Viroacfa4382008-12-04 10:06:33 -05002917 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 return -EPERM;
2919
2920 error = security_inode_symlink(dir, dentry, oldname);
2921 if (error)
2922 return error;
2923
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002925 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002926 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 return error;
2928}
2929
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002930SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2931 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932{
Al Viro2ad94ae2008-07-21 09:32:51 -04002933 int error;
2934 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07002935 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002936 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937
2938 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002939 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002941
Al Virodae6ad82011-06-26 11:50:15 -04002942 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07002943 error = PTR_ERR(dentry);
2944 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04002945 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07002946
Al Virodae6ad82011-06-26 11:50:15 -04002947 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002948 if (error)
2949 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002950 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002951 if (error)
2952 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04002953 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002954out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002955 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08002956out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002957 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002958 mutex_unlock(&path.dentry->d_inode->i_mutex);
2959 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07002960out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 putname(from);
2962 return error;
2963}
2964
Heiko Carstens3480b252009-01-14 14:14:16 +01002965SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002966{
2967 return sys_symlinkat(oldname, AT_FDCWD, newname);
2968}
2969
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2971{
2972 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05002973 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 int error;
2975
2976 if (!inode)
2977 return -ENOENT;
2978
Miklos Szeredia95164d2008-07-30 15:08:48 +02002979 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 if (error)
2981 return error;
2982
2983 if (dir->i_sb != inode->i_sb)
2984 return -EXDEV;
2985
2986 /*
2987 * A link to an append-only or immutable file cannot be created.
2988 */
2989 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2990 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002991 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002993 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 return -EPERM;
2995
2996 error = security_inode_link(old_dentry, dir, new_dentry);
2997 if (error)
2998 return error;
2999
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003000 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303001 /* Make sure we don't allow creating hardlink to an unlinked file */
3002 if (inode->i_nlink == 0)
3003 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003004 else if (max_links && inode->i_nlink >= max_links)
3005 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303006 else
3007 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003008 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003009 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003010 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 return error;
3012}
3013
3014/*
3015 * Hardlinks are often used in delicate situations. We avoid
3016 * security-related surprises by not following symlinks on the
3017 * newname. --KAB
3018 *
3019 * We don't follow them on the oldname either to be compatible
3020 * with linux 2.0, and to avoid hard-linking to directories
3021 * and other special files. --ADM
3022 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003023SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3024 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025{
3026 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003027 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303028 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303031 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003032 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303033 /*
3034 * To use null names we require CAP_DAC_READ_SEARCH
3035 * This ensures that not everyone will be able to create
3036 * handlink using the passed filedescriptor.
3037 */
3038 if (flags & AT_EMPTY_PATH) {
3039 if (!capable(CAP_DAC_READ_SEARCH))
3040 return -ENOENT;
3041 how = LOOKUP_EMPTY;
3042 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003043
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303044 if (flags & AT_SYMLINK_FOLLOW)
3045 how |= LOOKUP_FOLLOW;
3046
3047 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003049 return error;
3050
Al Virodae6ad82011-06-26 11:50:15 -04003051 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003053 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003054 goto out;
3055
3056 error = -EXDEV;
3057 if (old_path.mnt != new_path.mnt)
3058 goto out_dput;
3059 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003060 if (error)
3061 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003062 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003063 if (error)
3064 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003065 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003066out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003067 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003068out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003069 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003070 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3071 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072out:
Al Viro2d8f3032008-07-22 09:59:21 -04003073 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074
3075 return error;
3076}
3077
Heiko Carstens3480b252009-01-14 14:14:16 +01003078SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003079{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003080 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003081}
3082
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083/*
3084 * The worst of all namespace operations - renaming directory. "Perverted"
3085 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3086 * Problems:
3087 * a) we can get into loop creation. Check is done in is_subdir().
3088 * b) race potential - two innocent renames can create a loop together.
3089 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003090 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 * story.
3092 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003093 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 * whether the target exists). Solution: try to be smart with locking
3095 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003096 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 * move will be locked. Thus we can rank directories by the tree
3098 * (ancestors first) and rank all non-directories after them.
3099 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003100 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 * HOWEVER, it relies on the assumption that any object with ->lookup()
3102 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3103 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003104 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003105 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003107 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 * locking].
3109 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003110static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3111 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112{
3113 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003114 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003115 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116
3117 /*
3118 * If we are going to change the parent - check write permissions,
3119 * we'll need to flip '..'.
3120 */
3121 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003122 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 if (error)
3124 return error;
3125 }
3126
3127 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3128 if (error)
3129 return error;
3130
Al Viro1d2ef592011-09-14 18:55:41 +01003131 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003132 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003133 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003134
3135 error = -EBUSY;
3136 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3137 goto out;
3138
Al Viro8de52772012-02-06 12:45:27 -05003139 error = -EMLINK;
3140 if (max_links && !target && new_dir != old_dir &&
3141 new_dir->i_nlink >= max_links)
3142 goto out;
3143
Sage Weil3cebde22011-05-29 21:20:59 -07003144 if (target)
3145 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003146 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3147 if (error)
3148 goto out;
3149
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003151 target->i_flags |= S_DEAD;
3152 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 }
Sage Weil9055cba2011-05-24 13:06:12 -07003154out:
3155 if (target)
3156 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003157 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003158 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003159 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3160 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 return error;
3162}
3163
Adrian Bunk75c96f82005-05-05 16:16:09 -07003164static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3165 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166{
Sage Weil51892bb2011-05-24 13:06:13 -07003167 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 int error;
3169
3170 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3171 if (error)
3172 return error;
3173
3174 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003176 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003177
3178 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003180 goto out;
3181
3182 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3183 if (error)
3184 goto out;
3185
3186 if (target)
3187 dont_mount(new_dentry);
3188 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3189 d_move(old_dentry, new_dentry);
3190out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003192 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 dput(new_dentry);
3194 return error;
3195}
3196
3197int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3198 struct inode *new_dir, struct dentry *new_dentry)
3199{
3200 int error;
3201 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003202 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203
3204 if (old_dentry->d_inode == new_dentry->d_inode)
3205 return 0;
3206
3207 error = may_delete(old_dir, old_dentry, is_dir);
3208 if (error)
3209 return error;
3210
3211 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003212 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 else
3214 error = may_delete(new_dir, new_dentry, is_dir);
3215 if (error)
3216 return error;
3217
Al Viroacfa4382008-12-04 10:06:33 -05003218 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 return -EPERM;
3220
Robert Love0eeca282005-07-12 17:06:03 -04003221 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3222
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 if (is_dir)
3224 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3225 else
3226 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003227 if (!error)
3228 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003229 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003230 fsnotify_oldname_free(old_name);
3231
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 return error;
3233}
3234
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003235SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3236 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237{
Al Viro2ad94ae2008-07-21 09:32:51 -04003238 struct dentry *old_dir, *new_dir;
3239 struct dentry *old_dentry, *new_dentry;
3240 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003242 char *from;
3243 char *to;
3244 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245
Al Viro2ad94ae2008-07-21 09:32:51 -04003246 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247 if (error)
3248 goto exit;
3249
Al Viro2ad94ae2008-07-21 09:32:51 -04003250 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 if (error)
3252 goto exit1;
3253
3254 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003255 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 goto exit2;
3257
Jan Blunck4ac91372008-02-14 19:34:32 -08003258 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 error = -EBUSY;
3260 if (oldnd.last_type != LAST_NORM)
3261 goto exit2;
3262
Jan Blunck4ac91372008-02-14 19:34:32 -08003263 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 if (newnd.last_type != LAST_NORM)
3265 goto exit2;
3266
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003267 oldnd.flags &= ~LOOKUP_PARENT;
3268 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003269 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003270
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 trap = lock_rename(new_dir, old_dir);
3272
Christoph Hellwig49705b72005-11-08 21:35:06 -08003273 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 error = PTR_ERR(old_dentry);
3275 if (IS_ERR(old_dentry))
3276 goto exit3;
3277 /* source must exist */
3278 error = -ENOENT;
3279 if (!old_dentry->d_inode)
3280 goto exit4;
3281 /* unless the source is a directory trailing slashes give -ENOTDIR */
3282 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3283 error = -ENOTDIR;
3284 if (oldnd.last.name[oldnd.last.len])
3285 goto exit4;
3286 if (newnd.last.name[newnd.last.len])
3287 goto exit4;
3288 }
3289 /* source should not be ancestor of target */
3290 error = -EINVAL;
3291 if (old_dentry == trap)
3292 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003293 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 error = PTR_ERR(new_dentry);
3295 if (IS_ERR(new_dentry))
3296 goto exit4;
3297 /* target should not be an ancestor of source */
3298 error = -ENOTEMPTY;
3299 if (new_dentry == trap)
3300 goto exit5;
3301
Dave Hansen9079b1e2008-02-15 14:37:49 -08003302 error = mnt_want_write(oldnd.path.mnt);
3303 if (error)
3304 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003305 error = security_path_rename(&oldnd.path, old_dentry,
3306 &newnd.path, new_dentry);
3307 if (error)
3308 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 error = vfs_rename(old_dir->d_inode, old_dentry,
3310 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003311exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003312 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313exit5:
3314 dput(new_dentry);
3315exit4:
3316 dput(old_dentry);
3317exit3:
3318 unlock_rename(new_dir, old_dir);
3319exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003320 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003321 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003323 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003325exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 return error;
3327}
3328
Heiko Carstensa26eab22009-01-14 14:14:17 +01003329SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003330{
3331 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3332}
3333
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3335{
3336 int len;
3337
3338 len = PTR_ERR(link);
3339 if (IS_ERR(link))
3340 goto out;
3341
3342 len = strlen(link);
3343 if (len > (unsigned) buflen)
3344 len = buflen;
3345 if (copy_to_user(buffer, link, len))
3346 len = -EFAULT;
3347out:
3348 return len;
3349}
3350
3351/*
3352 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3353 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3354 * using) it for any given inode is up to filesystem.
3355 */
3356int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3357{
3358 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003359 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003360 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003361
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003363 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003364 if (IS_ERR(cookie))
3365 return PTR_ERR(cookie);
3366
3367 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3368 if (dentry->d_inode->i_op->put_link)
3369 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3370 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371}
3372
3373int vfs_follow_link(struct nameidata *nd, const char *link)
3374{
3375 return __vfs_follow_link(nd, link);
3376}
3377
3378/* get the link contents into pagecache */
3379static char *page_getlink(struct dentry * dentry, struct page **ppage)
3380{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003381 char *kaddr;
3382 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003384 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003386 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003388 kaddr = kmap(page);
3389 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3390 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391}
3392
3393int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3394{
3395 struct page *page = NULL;
3396 char *s = page_getlink(dentry, &page);
3397 int res = vfs_readlink(dentry,buffer,buflen,s);
3398 if (page) {
3399 kunmap(page);
3400 page_cache_release(page);
3401 }
3402 return res;
3403}
3404
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003405void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003407 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003409 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410}
3411
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003412void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003414 struct page *page = cookie;
3415
3416 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 kunmap(page);
3418 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 }
3420}
3421
Nick Piggin54566b22009-01-04 12:00:53 -08003422/*
3423 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3424 */
3425int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426{
3427 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003428 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003429 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003430 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003432 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3433 if (nofs)
3434 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435
NeilBrown7e53cac2006-03-25 03:07:57 -08003436retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003437 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003438 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003440 goto fail;
3441
Cong Wange8e3c3d2011-11-25 23:14:27 +08003442 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08003444 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07003445
3446 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3447 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448 if (err < 0)
3449 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003450 if (err < len-1)
3451 goto retry;
3452
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 mark_inode_dirty(inode);
3454 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455fail:
3456 return err;
3457}
3458
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003459int page_symlink(struct inode *inode, const char *symname, int len)
3460{
3461 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003462 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003463}
3464
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003465const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 .readlink = generic_readlink,
3467 .follow_link = page_follow_link_light,
3468 .put_link = page_put_link,
3469};
3470
Al Viro2d8f3032008-07-22 09:59:21 -04003471EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003472EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473EXPORT_SYMBOL(follow_down);
3474EXPORT_SYMBOL(follow_up);
3475EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3476EXPORT_SYMBOL(getname);
3477EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478EXPORT_SYMBOL(lookup_one_len);
3479EXPORT_SYMBOL(page_follow_link_light);
3480EXPORT_SYMBOL(page_put_link);
3481EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003482EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483EXPORT_SYMBOL(page_symlink);
3484EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003485EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003486EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003487EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488EXPORT_SYMBOL(unlock_rename);
3489EXPORT_SYMBOL(vfs_create);
3490EXPORT_SYMBOL(vfs_follow_link);
3491EXPORT_SYMBOL(vfs_link);
3492EXPORT_SYMBOL(vfs_mkdir);
3493EXPORT_SYMBOL(vfs_mknod);
3494EXPORT_SYMBOL(generic_permission);
3495EXPORT_SYMBOL(vfs_readlink);
3496EXPORT_SYMBOL(vfs_rename);
3497EXPORT_SYMBOL(vfs_rmdir);
3498EXPORT_SYMBOL(vfs_symlink);
3499EXPORT_SYMBOL(vfs_unlink);
3500EXPORT_SYMBOL(dentry_unhash);
3501EXPORT_SYMBOL(generic_readlink);