blob: 3cb616d38d9cf616f3b381c34b988f5d37f6fac2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
Eric Parise81e3f42009-12-04 15:47:36 -050037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800117static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
Al Virof52e0c12011-03-14 18:56:51 -0400139static char *getname_flags(const char __user * filename, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
Al Virof52e0c12011-03-14 18:56:51 -0400150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
Al Virof52e0c12011-03-14 18:56:51 -0400160char *getname(const char __user * filename)
161{
162 return getname_flags(filename, 0);
163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165#ifdef CONFIG_AUDITSYSCALL
166void putname(const char *name)
167{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400168 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 audit_putname(name);
170 else
171 __putname(name);
172}
173EXPORT_SYMBOL(putname);
174#endif
175
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700176/*
177 * This does basic POSIX ACL permission checking
178 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100179static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700181{
182 umode_t mode = inode->i_mode;
183
184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
185
Serge E. Hallyne795b712011-03-23 16:43:25 -0700186 if (current_user_ns() != inode_userns(inode))
187 goto other_perms;
188
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700189 if (current_fsuid() == inode->i_uid)
190 mode >>= 6;
191 else {
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100193 int error = check_acl(inode, mask, flags);
194 if (error != -EAGAIN)
195 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
Serge E. Hallyne795b712011-03-23 16:43:25 -0700202other_perms:
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700203 /*
204 * If the DACs are ok we don't need any capability check.
205 */
206 if ((mask & ~mode) == 0)
207 return 0;
208 return -EACCES;
209}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100212 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
Randy Dunlap39191622011-01-08 19:36:21 -0800216 * @flags: IPERM_FLAG_ flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 *
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100221 * are used for other things.
222 *
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100227int generic_permission(struct inode *inode, int mask, unsigned int flags,
228 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700230 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700233 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100235 ret = acl_permission_check(inode, mask, flags, check_acl);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700236 if (ret != -EACCES)
237 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 /*
240 * Read/write DACs are always overridable.
241 * Executable DACs are overridable if at least one exec bit is set.
242 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200243 if (!(mask & MAY_EXEC) || execute_ok(inode))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700244 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 return 0;
246
247 /*
248 * Searching includes executable on directories, else just read.
249 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600250 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700252 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 return 0;
254
255 return -EACCES;
256}
257
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200258/**
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
262 *
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
267 */
Al Virof419a2e2008-07-22 00:07:17 -0400268int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Al Viroe6305c42008-07-15 21:03:57 -0400270 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
272 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700273 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 /*
276 * Nobody gets write access to a read-only fs.
277 */
278 if (IS_RDONLY(inode) &&
279 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
280 return -EROFS;
281
282 /*
283 * Nobody gets write access to an immutable file.
284 */
285 if (IS_IMMUTABLE(inode))
286 return -EACCES;
287 }
288
Al Viroacfa4382008-12-04 10:06:33 -0500289 if (inode->i_op->permission)
Nick Pigginb74c79e2011-01-07 17:49:58 +1100290 retval = inode->i_op->permission(inode, mask, 0);
Miklos Szeredif696a362008-07-31 13:41:58 +0200291 else
Nick Pigginb74c79e2011-01-07 17:49:58 +1100292 retval = generic_permission(inode, mask, 0,
293 inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 if (retval)
296 return retval;
297
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700298 retval = devcgroup_inode_permission(inode, mask);
299 if (retval)
300 return retval;
301
Eric Parisd09ca732010-07-23 11:43:57 -0400302 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303}
304
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800305/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800306 * file_permission - check for additional access rights to a given file
307 * @file: file to check access rights for
308 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
309 *
310 * Used to check for read/write/execute permissions on an already opened
311 * file.
312 *
313 * Note:
314 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200315 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800316 */
317int file_permission(struct file *file, int mask)
318{
Al Virof419a2e2008-07-22 00:07:17 -0400319 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800320}
321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322/*
323 * get_write_access() gets write permission for a file.
324 * put_write_access() releases this write permission.
325 * This is used for regular files.
326 * We cannot support write (and maybe mmap read-write shared) accesses and
327 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
328 * can have the following values:
329 * 0: no writers, no VM_DENYWRITE mappings
330 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
331 * > 0: (i_writecount) users are writing to the file.
332 *
333 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
334 * except for the cases where we don't hold i_writecount yet. Then we need to
335 * use {get,deny}_write_access() - these functions check the sign and refuse
336 * to do the change if sign is wrong. Exclusion between them is provided by
337 * the inode->i_lock spinlock.
338 */
339
340int get_write_access(struct inode * inode)
341{
342 spin_lock(&inode->i_lock);
343 if (atomic_read(&inode->i_writecount) < 0) {
344 spin_unlock(&inode->i_lock);
345 return -ETXTBSY;
346 }
347 atomic_inc(&inode->i_writecount);
348 spin_unlock(&inode->i_lock);
349
350 return 0;
351}
352
353int deny_write_access(struct file * file)
354{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800355 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
357 spin_lock(&inode->i_lock);
358 if (atomic_read(&inode->i_writecount) > 0) {
359 spin_unlock(&inode->i_lock);
360 return -ETXTBSY;
361 }
362 atomic_dec(&inode->i_writecount);
363 spin_unlock(&inode->i_lock);
364
365 return 0;
366}
367
Jan Blunck1d957f92008-02-14 19:34:35 -0800368/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800369 * path_get - get a reference to a path
370 * @path: path to get the reference to
371 *
372 * Given a path increment the reference count to the dentry and the vfsmount.
373 */
374void path_get(struct path *path)
375{
376 mntget(path->mnt);
377 dget(path->dentry);
378}
379EXPORT_SYMBOL(path_get);
380
381/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800382 * path_put - put a reference to a path
383 * @path: path to put the reference to
384 *
385 * Given a path decrement the reference count to the dentry and the vfsmount.
386 */
387void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Jan Blunck1d957f92008-02-14 19:34:35 -0800389 dput(path->dentry);
390 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
Jan Blunck1d957f92008-02-14 19:34:35 -0800392EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
Trond Myklebust834f2a42005-10-18 14:20:16 -0700394/**
Nick Piggin31e6b012011-01-07 17:49:52 +1100395 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
396 * @nd: nameidata pathwalk data to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800397 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100398 *
399 * Path walking has 2 modes, rcu-walk and ref-walk (see
400 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
401 * to drop out of rcu-walk mode and take normal reference counts on dentries
402 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
403 * refcounts at the last known good point before rcu-walk got stuck, so
404 * ref-walk may continue from there. If this is not successful (eg. a seqcount
405 * has changed), then failure is returned and path walk restarts from the
406 * beginning in ref-walk mode.
407 *
408 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
409 * ref-walk. Must be called from rcu-walk context.
410 */
411static int nameidata_drop_rcu(struct nameidata *nd)
412{
413 struct fs_struct *fs = current->fs;
414 struct dentry *dentry = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500415 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100416
417 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500418 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
419 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100420 spin_lock(&fs->lock);
421 if (nd->root.mnt != fs->root.mnt ||
422 nd->root.dentry != fs->root.dentry)
423 goto err_root;
424 }
425 spin_lock(&dentry->d_lock);
426 if (!__d_rcu_to_refcount(dentry, nd->seq))
427 goto err;
428 BUG_ON(nd->inode != dentry->d_inode);
429 spin_unlock(&dentry->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500430 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100431 path_get(&nd->root);
432 spin_unlock(&fs->lock);
433 }
434 mntget(nd->path.mnt);
435
436 rcu_read_unlock();
437 br_read_unlock(vfsmount_lock);
438 nd->flags &= ~LOOKUP_RCU;
439 return 0;
440err:
441 spin_unlock(&dentry->d_lock);
442err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500443 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100444 spin_unlock(&fs->lock);
445 return -ECHILD;
446}
447
448/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
449static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
450{
451 if (nd->flags & LOOKUP_RCU)
452 return nameidata_drop_rcu(nd);
453 return 0;
454}
455
456/**
457 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
458 * @nd: nameidata pathwalk data to drop
459 * @dentry: dentry to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800460 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100461 *
462 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
463 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
464 * @nd. Must be called from rcu-walk context.
465 */
466static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
467{
468 struct fs_struct *fs = current->fs;
469 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500470 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100471
472 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500473 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
474 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100475 spin_lock(&fs->lock);
476 if (nd->root.mnt != fs->root.mnt ||
477 nd->root.dentry != fs->root.dentry)
478 goto err_root;
479 }
480 spin_lock(&parent->d_lock);
481 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
482 if (!__d_rcu_to_refcount(dentry, nd->seq))
483 goto err;
484 /*
485 * If the sequence check on the child dentry passed, then the child has
486 * not been removed from its parent. This means the parent dentry must
487 * be valid and able to take a reference at this point.
488 */
489 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
490 BUG_ON(!parent->d_count);
491 parent->d_count++;
492 spin_unlock(&dentry->d_lock);
493 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500494 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100495 path_get(&nd->root);
496 spin_unlock(&fs->lock);
497 }
498 mntget(nd->path.mnt);
499
500 rcu_read_unlock();
501 br_read_unlock(vfsmount_lock);
502 nd->flags &= ~LOOKUP_RCU;
503 return 0;
504err:
505 spin_unlock(&dentry->d_lock);
506 spin_unlock(&parent->d_lock);
507err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500508 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100509 spin_unlock(&fs->lock);
510 return -ECHILD;
511}
512
513/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
514static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
515{
Al Viroa7472ba2011-03-04 14:39:30 -0500516 if (nd->flags & LOOKUP_RCU) {
517 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
518 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500519 if (!(nd->flags & LOOKUP_ROOT))
520 nd->root.mnt = NULL;
Al Viroa7472ba2011-03-04 14:39:30 -0500521 rcu_read_unlock();
522 br_read_unlock(vfsmount_lock);
523 return -ECHILD;
524 }
525 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100526 return 0;
527}
528
529/**
530 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
531 * @nd: nameidata pathwalk data to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800532 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100533 *
534 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
535 * nd->path should be the final element of the lookup, so nd->root is discarded.
536 * Must be called from rcu-walk context.
537 */
538static int nameidata_drop_rcu_last(struct nameidata *nd)
539{
540 struct dentry *dentry = nd->path.dentry;
541
542 BUG_ON(!(nd->flags & LOOKUP_RCU));
543 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500544 if (!(nd->flags & LOOKUP_ROOT))
545 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100546 spin_lock(&dentry->d_lock);
547 if (!__d_rcu_to_refcount(dentry, nd->seq))
548 goto err_unlock;
549 BUG_ON(nd->inode != dentry->d_inode);
550 spin_unlock(&dentry->d_lock);
551
552 mntget(nd->path.mnt);
553
554 rcu_read_unlock();
555 br_read_unlock(vfsmount_lock);
556
557 return 0;
558
559err_unlock:
560 spin_unlock(&dentry->d_lock);
561 rcu_read_unlock();
562 br_read_unlock(vfsmount_lock);
563 return -ECHILD;
564}
565
Nick Piggin31e6b012011-01-07 17:49:52 +1100566/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700567 * release_open_intent - free up open intent resources
568 * @nd: pointer to nameidata
569 */
570void release_open_intent(struct nameidata *nd)
571{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800572 struct file *file = nd->intent.open.file;
573
574 if (file && !IS_ERR(file)) {
575 if (file->f_path.dentry == NULL)
576 put_filp(file);
577 else
578 fput(file);
579 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700580}
581
Al Virof60aef72011-02-15 01:35:28 -0500582static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100583{
Al Virof60aef72011-02-15 01:35:28 -0500584 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100585}
586
Al Virof5e1c1c2011-02-15 01:32:55 -0500587static struct dentry *
Ian Kentbcdc5e02006-09-27 01:50:44 -0700588do_revalidate(struct dentry *dentry, struct nameidata *nd)
589{
Al Virof5e1c1c2011-02-15 01:32:55 -0500590 int status = d_revalidate(dentry, nd);
Ian Kentbcdc5e02006-09-27 01:50:44 -0700591 if (unlikely(status <= 0)) {
592 /*
593 * The dentry failed validation.
594 * If d_revalidate returned 0 attempt to invalidate
595 * the dentry otherwise d_revalidate is asking us
596 * to return a fail status.
597 */
Nick Piggin34286d62011-01-07 17:49:57 +1100598 if (status < 0) {
Al Virof5e1c1c2011-02-15 01:32:55 -0500599 dput(dentry);
Nick Piggin34286d62011-01-07 17:49:57 +1100600 dentry = ERR_PTR(status);
Al Virof5e1c1c2011-02-15 01:32:55 -0500601 } else if (!d_invalidate(dentry)) {
602 dput(dentry);
603 dentry = NULL;
Ian Kentbcdc5e02006-09-27 01:50:44 -0700604 }
605 }
606 return dentry;
607}
608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609/*
Al Viro16c2cd72011-02-22 15:50:10 -0500610 * handle_reval_path - force revalidation of a dentry
Jeff Layton39159de2009-12-07 12:01:50 -0500611 *
612 * In some situations the path walking code will trust dentries without
613 * revalidating them. This causes problems for filesystems that depend on
614 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
615 * (which indicates that it's possible for the dentry to go stale), force
616 * a d_revalidate call before proceeding.
617 *
618 * Returns 0 if the revalidation was successful. If the revalidation fails,
619 * either return the error returned by d_revalidate or -ESTALE if the
620 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
621 * invalidate the dentry. It's up to the caller to handle putting references
622 * to the path if necessary.
623 */
Al Viro16c2cd72011-02-22 15:50:10 -0500624static inline int handle_reval_path(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500625{
Al Viro16c2cd72011-02-22 15:50:10 -0500626 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500627 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500628
Al Viro16c2cd72011-02-22 15:50:10 -0500629 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500630 return 0;
631
Al Viro16c2cd72011-02-22 15:50:10 -0500632 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
633 return 0;
634
635 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
636 return 0;
637
638 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100639 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500640 if (status > 0)
641 return 0;
642
Al Viro16c2cd72011-02-22 15:50:10 -0500643 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500644 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500645
Jeff Layton39159de2009-12-07 12:01:50 -0500646 return status;
647}
648
649/*
Al Virob75b5082009-12-16 01:01:38 -0500650 * Short-cut version of permission(), for calling on directories
651 * during pathname resolution. Combines parts of permission()
652 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 *
654 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500655 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 * complete permission check.
657 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100658static inline int exec_permission(struct inode *inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700660 int ret;
Serge E. Hallyne795b712011-03-23 16:43:25 -0700661 struct user_namespace *ns = inode_userns(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700663 if (inode->i_op->permission) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100664 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
665 } else {
666 ret = acl_permission_check(inode, MAY_EXEC, flags,
667 inode->i_op->check_acl);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700668 }
Nick Pigginb74c79e2011-01-07 17:49:58 +1100669 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 goto ok;
Nick Pigginb74c79e2011-01-07 17:49:58 +1100671 if (ret == -ECHILD)
Nick Piggin31e6b012011-01-07 17:49:52 +1100672 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Serge E. Hallyne795b712011-03-23 16:43:25 -0700674 if (ns_capable(ns, CAP_DAC_OVERRIDE) ||
675 ns_capable(ns, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 goto ok;
677
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700678 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679ok:
Nick Pigginb74c79e2011-01-07 17:49:58 +1100680 return security_inode_exec_permission(inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
Al Viro2a737872009-04-07 11:49:53 -0400683static __always_inline void set_root(struct nameidata *nd)
684{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200685 if (!nd->root.mnt)
686 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400687}
688
Al Viro6de88d72009-08-09 01:41:57 +0400689static int link_path_walk(const char *, struct nameidata *);
690
Nick Piggin31e6b012011-01-07 17:49:52 +1100691static __always_inline void set_root_rcu(struct nameidata *nd)
692{
693 if (!nd->root.mnt) {
694 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100695 unsigned seq;
696
697 do {
698 seq = read_seqcount_begin(&fs->seq);
699 nd->root = fs->root;
700 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100701 }
702}
703
Arjan van de Venf1662352006-01-14 13:21:31 -0800704static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Nick Piggin31e6b012011-01-07 17:49:52 +1100706 int ret;
707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (IS_ERR(link))
709 goto fail;
710
711 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400712 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800713 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400714 nd->path = nd->root;
715 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500716 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100718 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100719
Nick Piggin31e6b012011-01-07 17:49:52 +1100720 ret = link_path_walk(link, nd);
721 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800723 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return PTR_ERR(link);
725}
726
Jan Blunck1d957f92008-02-14 19:34:35 -0800727static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700728{
729 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800730 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700731 mntput(path->mnt);
732}
733
Nick Piggin7b9337a2011-01-14 08:42:43 +0000734static inline void path_to_nameidata(const struct path *path,
735 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700736{
Nick Piggin31e6b012011-01-07 17:49:52 +1100737 if (!(nd->flags & LOOKUP_RCU)) {
738 dput(nd->path.dentry);
739 if (nd->path.mnt != path->mnt)
740 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800741 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100742 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800743 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700744}
745
Al Viro574197e2011-03-14 22:20:34 -0400746static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
747{
748 struct inode *inode = link->dentry->d_inode;
749 if (!IS_ERR(cookie) && inode->i_op->put_link)
750 inode->i_op->put_link(link->dentry, nd, cookie);
751 path_put(link);
752}
753
Al Virodef4af32009-12-26 08:37:05 -0500754static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400755follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800756{
757 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000758 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800759
Al Viro844a3912011-02-15 00:38:26 -0500760 BUG_ON(nd->flags & LOOKUP_RCU);
761
Al Viro0e794582011-03-16 02:45:02 -0400762 if (link->mnt == nd->path.mnt)
763 mntget(link->mnt);
764
Al Viro574197e2011-03-14 22:20:34 -0400765 if (unlikely(current->total_link_count >= 40)) {
766 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400767 path_put(&nd->path);
768 return -ELOOP;
769 }
770 cond_resched();
771 current->total_link_count++;
772
Nick Piggin7b9337a2011-01-14 08:42:43 +0000773 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800774 nd_set_link(nd, NULL);
775
Al Viro36f3b4f2011-02-22 21:24:38 -0500776 error = security_inode_follow_link(link->dentry, nd);
777 if (error) {
778 *p = ERR_PTR(error); /* no ->put_link(), please */
779 path_put(&nd->path);
780 return error;
781 }
782
Al Viro86acdca12009-12-22 23:45:11 -0500783 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500784 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
785 error = PTR_ERR(*p);
786 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800787 char *s = nd_get_link(nd);
788 error = 0;
789 if (s)
790 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400791 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500792 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400793 nd->inode = nd->path.dentry->d_inode;
794 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400795 /* stepped on a _really_ weird one */
796 path_put(&nd->path);
797 error = -ELOOP;
798 }
799 }
Ian Kent051d3812006-03-27 01:14:53 -0800800 }
Ian Kent051d3812006-03-27 01:14:53 -0800801 return error;
802}
803
Nick Piggin31e6b012011-01-07 17:49:52 +1100804static int follow_up_rcu(struct path *path)
805{
806 struct vfsmount *parent;
807 struct dentry *mountpoint;
808
809 parent = path->mnt->mnt_parent;
810 if (parent == path->mnt)
811 return 0;
812 mountpoint = path->mnt->mnt_mountpoint;
813 path->dentry = mountpoint;
814 path->mnt = parent;
815 return 1;
816}
817
Al Virobab77eb2009-04-18 03:26:48 -0400818int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
820 struct vfsmount *parent;
821 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000822
823 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400824 parent = path->mnt->mnt_parent;
825 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000826 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return 0;
828 }
829 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400830 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000831 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400832 dput(path->dentry);
833 path->dentry = mountpoint;
834 mntput(path->mnt);
835 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 return 1;
837}
838
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100839/*
David Howells9875cf82011-01-14 18:45:21 +0000840 * Perform an automount
841 * - return -EISDIR to tell follow_managed() to stop and return the path we
842 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
David Howells9875cf82011-01-14 18:45:21 +0000844static int follow_automount(struct path *path, unsigned flags,
845 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100846{
David Howells9875cf82011-01-14 18:45:21 +0000847 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000848 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100849
David Howells9875cf82011-01-14 18:45:21 +0000850 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
851 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700852
David Howells6f45b652011-01-14 18:45:31 +0000853 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
854 * and this is the terminal part of the path.
855 */
856 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
857 return -EISDIR; /* we actually want to stop here */
858
David Howells9875cf82011-01-14 18:45:21 +0000859 /* We want to mount if someone is trying to open/create a file of any
860 * type under the mountpoint, wants to traverse through the mountpoint
861 * or wants to open the mounted directory.
862 *
863 * We don't want to mount if someone's just doing a stat and they've
864 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
865 * appended a '/' to the name.
866 */
867 if (!(flags & LOOKUP_FOLLOW) &&
868 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
869 LOOKUP_OPEN | LOOKUP_CREATE)))
870 return -EISDIR;
871
872 current->total_link_count++;
873 if (current->total_link_count >= 40)
874 return -ELOOP;
875
876 mnt = path->dentry->d_op->d_automount(path);
877 if (IS_ERR(mnt)) {
878 /*
879 * The filesystem is allowed to return -EISDIR here to indicate
880 * it doesn't want to automount. For instance, autofs would do
881 * this so that its userspace daemon can mount on this dentry.
882 *
883 * However, we can only permit this if it's a terminal point in
884 * the path being looked up; if it wasn't then the remainder of
885 * the path is inaccessible and we should say so.
886 */
887 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
888 return -EREMOTE;
889 return PTR_ERR(mnt);
890 }
David Howellsea5b7782011-01-14 19:10:03 +0000891
David Howells9875cf82011-01-14 18:45:21 +0000892 if (!mnt) /* mount collision */
893 return 0;
894
Al Viro19a167a2011-01-17 01:35:23 -0500895 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000896
David Howellsea5b7782011-01-14 19:10:03 +0000897 switch (err) {
898 case -EBUSY:
899 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500900 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000901 case 0:
David Howellsea5b7782011-01-14 19:10:03 +0000902 dput(path->dentry);
903 if (*need_mntput)
904 mntput(path->mnt);
905 path->mnt = mnt;
906 path->dentry = dget(mnt->mnt_root);
907 *need_mntput = true;
908 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500909 default:
910 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000911 }
Al Viro19a167a2011-01-17 01:35:23 -0500912
David Howells9875cf82011-01-14 18:45:21 +0000913}
914
915/*
916 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000917 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000918 * - Flagged as mountpoint
919 * - Flagged as automount point
920 *
921 * This may only be called in refwalk mode.
922 *
923 * Serialization is taken care of in namespace.c
924 */
925static int follow_managed(struct path *path, unsigned flags)
926{
927 unsigned managed;
928 bool need_mntput = false;
929 int ret;
930
931 /* Given that we're not holding a lock here, we retain the value in a
932 * local variable for each dentry as we look at it so that we don't see
933 * the components of that value change under us */
934 while (managed = ACCESS_ONCE(path->dentry->d_flags),
935 managed &= DCACHE_MANAGED_DENTRY,
936 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000937 /* Allow the filesystem to manage the transit without i_mutex
938 * being held. */
939 if (managed & DCACHE_MANAGE_TRANSIT) {
940 BUG_ON(!path->dentry->d_op);
941 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400942 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000943 if (ret < 0)
944 return ret == -EISDIR ? 0 : ret;
945 }
946
David Howells9875cf82011-01-14 18:45:21 +0000947 /* Transit to a mounted filesystem. */
948 if (managed & DCACHE_MOUNTED) {
949 struct vfsmount *mounted = lookup_mnt(path);
950 if (mounted) {
951 dput(path->dentry);
952 if (need_mntput)
953 mntput(path->mnt);
954 path->mnt = mounted;
955 path->dentry = dget(mounted->mnt_root);
956 need_mntput = true;
957 continue;
958 }
959
960 /* Something is mounted on this dentry in another
961 * namespace and/or whatever was mounted there in this
962 * namespace got unmounted before we managed to get the
963 * vfsmount_lock */
964 }
965
966 /* Handle an automount point */
967 if (managed & DCACHE_NEED_AUTOMOUNT) {
968 ret = follow_automount(path, flags, &need_mntput);
969 if (ret < 0)
970 return ret == -EISDIR ? 0 : ret;
971 continue;
972 }
973
974 /* We didn't change the current path point */
975 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 }
David Howells9875cf82011-01-14 18:45:21 +0000977 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978}
979
David Howellscc53ce52011-01-14 18:45:26 +0000980int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
982 struct vfsmount *mounted;
983
Al Viro1c755af2009-04-18 14:06:57 -0400984 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400986 dput(path->dentry);
987 mntput(path->mnt);
988 path->mnt = mounted;
989 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 return 1;
991 }
992 return 0;
993}
994
Ian Kent62a73752011-03-25 01:51:02 +0800995static inline bool managed_dentry_might_block(struct dentry *dentry)
996{
997 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
998 dentry->d_op->d_manage(dentry, true) < 0);
999}
1000
David Howells9875cf82011-01-14 18:45:21 +00001001/*
1002 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
David Howellscc53ce52011-01-14 18:45:26 +00001003 * meet a managed dentry and we're not walking to "..". True is returned to
David Howells9875cf82011-01-14 18:45:21 +00001004 * continue, false to abort.
1005 */
1006static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1007 struct inode **inode, bool reverse_transit)
1008{
Ian Kent62a73752011-03-25 01:51:02 +08001009 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +00001010 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001011 /*
1012 * Don't forget we might have a non-mountpoint managed dentry
1013 * that wants to block transit.
1014 */
1015 *inode = path->dentry->d_inode;
1016 if (!reverse_transit &&
1017 unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001018 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001019
1020 if (!d_mountpoint(path->dentry))
1021 break;
1022
David Howells9875cf82011-01-14 18:45:21 +00001023 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1024 if (!mounted)
1025 break;
1026 path->mnt = mounted;
1027 path->dentry = mounted->mnt_root;
1028 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
David Howells9875cf82011-01-14 18:45:21 +00001029 }
1030
1031 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1032 return reverse_transit;
1033 return true;
1034}
1035
Nick Piggin31e6b012011-01-07 17:49:52 +11001036static int follow_dotdot_rcu(struct nameidata *nd)
1037{
1038 struct inode *inode = nd->inode;
1039
1040 set_root_rcu(nd);
1041
David Howells9875cf82011-01-14 18:45:21 +00001042 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001043 if (nd->path.dentry == nd->root.dentry &&
1044 nd->path.mnt == nd->root.mnt) {
1045 break;
1046 }
1047 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1048 struct dentry *old = nd->path.dentry;
1049 struct dentry *parent = old->d_parent;
1050 unsigned seq;
1051
1052 seq = read_seqcount_begin(&parent->d_seq);
1053 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001054 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001055 inode = parent->d_inode;
1056 nd->path.dentry = parent;
1057 nd->seq = seq;
1058 break;
1059 }
1060 if (!follow_up_rcu(&nd->path))
1061 break;
1062 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1063 inode = nd->path.dentry->d_inode;
1064 }
David Howells9875cf82011-01-14 18:45:21 +00001065 __follow_mount_rcu(nd, &nd->path, &inode, true);
Nick Piggin31e6b012011-01-07 17:49:52 +11001066 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001067 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001068
1069failed:
1070 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001071 if (!(nd->flags & LOOKUP_ROOT))
1072 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -05001073 rcu_read_unlock();
1074 br_read_unlock(vfsmount_lock);
1075 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001076}
1077
David Howells9875cf82011-01-14 18:45:21 +00001078/*
David Howellscc53ce52011-01-14 18:45:26 +00001079 * Follow down to the covering mount currently visible to userspace. At each
1080 * point, the filesystem owning that dentry may be queried as to whether the
1081 * caller is permitted to proceed or not.
1082 *
1083 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1084 * being true).
1085 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001086int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001087{
1088 unsigned managed;
1089 int ret;
1090
1091 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1092 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1093 /* Allow the filesystem to manage the transit without i_mutex
1094 * being held.
1095 *
1096 * We indicate to the filesystem if someone is trying to mount
1097 * something here. This gives autofs the chance to deny anyone
1098 * other than its daemon the right to mount on its
1099 * superstructure.
1100 *
1101 * The filesystem may sleep at this point.
1102 */
1103 if (managed & DCACHE_MANAGE_TRANSIT) {
1104 BUG_ON(!path->dentry->d_op);
1105 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001106 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001107 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001108 if (ret < 0)
1109 return ret == -EISDIR ? 0 : ret;
1110 }
1111
1112 /* Transit to a mounted filesystem. */
1113 if (managed & DCACHE_MOUNTED) {
1114 struct vfsmount *mounted = lookup_mnt(path);
1115 if (!mounted)
1116 break;
1117 dput(path->dentry);
1118 mntput(path->mnt);
1119 path->mnt = mounted;
1120 path->dentry = dget(mounted->mnt_root);
1121 continue;
1122 }
1123
1124 /* Don't handle automount points here */
1125 break;
1126 }
1127 return 0;
1128}
1129
1130/*
David Howells9875cf82011-01-14 18:45:21 +00001131 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1132 */
1133static void follow_mount(struct path *path)
1134{
1135 while (d_mountpoint(path->dentry)) {
1136 struct vfsmount *mounted = lookup_mnt(path);
1137 if (!mounted)
1138 break;
1139 dput(path->dentry);
1140 mntput(path->mnt);
1141 path->mnt = mounted;
1142 path->dentry = dget(mounted->mnt_root);
1143 }
1144}
1145
Nick Piggin31e6b012011-01-07 17:49:52 +11001146static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147{
Al Viro2a737872009-04-07 11:49:53 -04001148 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001151 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Al Viro2a737872009-04-07 11:49:53 -04001153 if (nd->path.dentry == nd->root.dentry &&
1154 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 break;
1156 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001157 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001158 /* rare case of legitimate dget_parent()... */
1159 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 dput(old);
1161 break;
1162 }
Al Viro3088dd72010-01-30 15:47:29 -05001163 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 }
Al Viro79ed0222009-04-18 13:59:41 -04001166 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001167 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168}
1169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001171 * Allocate a dentry with name and parent, and perform a parent
1172 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1173 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1174 * have verified that no child exists while under i_mutex.
1175 */
1176static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1177 struct qstr *name, struct nameidata *nd)
1178{
1179 struct inode *inode = parent->d_inode;
1180 struct dentry *dentry;
1181 struct dentry *old;
1182
1183 /* Don't create child dentry for a dead directory. */
1184 if (unlikely(IS_DEADDIR(inode)))
1185 return ERR_PTR(-ENOENT);
1186
1187 dentry = d_alloc(parent, name);
1188 if (unlikely(!dentry))
1189 return ERR_PTR(-ENOMEM);
1190
1191 old = inode->i_op->lookup(inode, dentry, nd);
1192 if (unlikely(old)) {
1193 dput(dentry);
1194 dentry = old;
1195 }
1196 return dentry;
1197}
1198
1199/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 * It's more convoluted than I'd like it to be, but... it's still fairly
1201 * small and for now I'd prefer to have fast path as straight as possible.
1202 * It _is_ time-critical.
1203 */
1204static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001205 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
Jan Blunck4ac91372008-02-14 19:34:32 -08001207 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001208 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001209 int need_reval = 1;
1210 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001211 int err;
1212
Al Viro3cac2602009-08-13 18:27:43 +04001213 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001214 * Rename seqlock is not required here because in the off chance
1215 * of a false negative due to a concurrent rename, we're going to
1216 * do the non-racy lookup, below.
1217 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001218 if (nd->flags & LOOKUP_RCU) {
1219 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001220 *inode = nd->inode;
1221 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001222 if (!dentry)
1223 goto unlazy;
1224
Nick Piggin31e6b012011-01-07 17:49:52 +11001225 /* Memory barrier in read_seqcount_begin of child is enough */
1226 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1227 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001228 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001229
Al Viro24643082011-02-15 01:26:22 -05001230 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001231 status = d_revalidate(dentry, nd);
1232 if (unlikely(status <= 0)) {
1233 if (status != -ECHILD)
1234 need_reval = 0;
1235 goto unlazy;
1236 }
Al Viro24643082011-02-15 01:26:22 -05001237 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001238 path->mnt = mnt;
1239 path->dentry = dentry;
David Howells9875cf82011-01-14 18:45:21 +00001240 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1241 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001242unlazy:
1243 if (dentry) {
1244 if (nameidata_dentry_drop_rcu(nd, dentry))
1245 return -ECHILD;
1246 } else {
1247 if (nameidata_drop_rcu(nd))
1248 return -ECHILD;
1249 }
1250 } else {
1251 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001252 }
Al Viro5a18fff2011-03-11 04:44:53 -05001253
1254retry:
1255 if (unlikely(!dentry)) {
1256 struct inode *dir = parent->d_inode;
1257 BUG_ON(nd->inode != dir);
1258
1259 mutex_lock(&dir->i_mutex);
1260 dentry = d_lookup(parent, name);
1261 if (likely(!dentry)) {
1262 dentry = d_alloc_and_lookup(parent, name, nd);
1263 if (IS_ERR(dentry)) {
1264 mutex_unlock(&dir->i_mutex);
1265 return PTR_ERR(dentry);
1266 }
1267 /* known good */
1268 need_reval = 0;
1269 status = 1;
1270 }
1271 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001272 }
Al Viro5a18fff2011-03-11 04:44:53 -05001273 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1274 status = d_revalidate(dentry, nd);
1275 if (unlikely(status <= 0)) {
1276 if (status < 0) {
1277 dput(dentry);
1278 return status;
1279 }
1280 if (!d_invalidate(dentry)) {
1281 dput(dentry);
1282 dentry = NULL;
1283 need_reval = 1;
1284 goto retry;
1285 }
1286 }
1287
David Howells9875cf82011-01-14 18:45:21 +00001288 path->mnt = mnt;
1289 path->dentry = dentry;
1290 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001291 if (unlikely(err < 0)) {
1292 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001293 return err;
Ian Kent89312212011-01-18 12:06:10 +08001294 }
David Howells9875cf82011-01-14 18:45:21 +00001295 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
1298
Al Viro52094c82011-02-21 21:34:47 -05001299static inline int may_lookup(struct nameidata *nd)
1300{
1301 if (nd->flags & LOOKUP_RCU) {
1302 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1303 if (err != -ECHILD)
1304 return err;
1305 if (nameidata_drop_rcu(nd))
1306 return -ECHILD;
1307 }
1308 return exec_permission(nd->inode, 0);
1309}
1310
Al Viro9856fa12011-03-04 14:22:06 -05001311static inline int handle_dots(struct nameidata *nd, int type)
1312{
1313 if (type == LAST_DOTDOT) {
1314 if (nd->flags & LOOKUP_RCU) {
1315 if (follow_dotdot_rcu(nd))
1316 return -ECHILD;
1317 } else
1318 follow_dotdot(nd);
1319 }
1320 return 0;
1321}
1322
Al Viro951361f2011-03-04 14:44:37 -05001323static void terminate_walk(struct nameidata *nd)
1324{
1325 if (!(nd->flags & LOOKUP_RCU)) {
1326 path_put(&nd->path);
1327 } else {
1328 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001329 if (!(nd->flags & LOOKUP_ROOT))
1330 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001331 rcu_read_unlock();
1332 br_read_unlock(vfsmount_lock);
1333 }
1334}
1335
Al Viroce57dfc2011-03-13 19:58:58 -04001336static inline int walk_component(struct nameidata *nd, struct path *path,
1337 struct qstr *name, int type, int follow)
1338{
1339 struct inode *inode;
1340 int err;
1341 /*
1342 * "." and ".." are special - ".." especially so because it has
1343 * to be able to know about the current root directory and
1344 * parent relationships.
1345 */
1346 if (unlikely(type != LAST_NORM))
1347 return handle_dots(nd, type);
1348 err = do_lookup(nd, name, path, &inode);
1349 if (unlikely(err)) {
1350 terminate_walk(nd);
1351 return err;
1352 }
1353 if (!inode) {
1354 path_to_nameidata(path, nd);
1355 terminate_walk(nd);
1356 return -ENOENT;
1357 }
1358 if (unlikely(inode->i_op->follow_link) && follow) {
1359 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1360 return -ECHILD;
1361 BUG_ON(inode != path->dentry->d_inode);
1362 return 1;
1363 }
1364 path_to_nameidata(path, nd);
1365 nd->inode = inode;
1366 return 0;
1367}
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369/*
Al Virob3563792011-03-14 21:54:55 -04001370 * This limits recursive symlink follows to 8, while
1371 * limiting consecutive symlinks to 40.
1372 *
1373 * Without that kind of total limit, nasty chains of consecutive
1374 * symlinks can cause almost arbitrarily long lookups.
1375 */
1376static inline int nested_symlink(struct path *path, struct nameidata *nd)
1377{
1378 int res;
1379
1380 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1381 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1382 path_put_conditional(path, nd);
1383 path_put(&nd->path);
1384 return -ELOOP;
1385 }
1386
1387 nd->depth++;
1388 current->link_count++;
1389
1390 do {
1391 struct path link = *path;
1392 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001393
1394 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001395 if (!res)
1396 res = walk_component(nd, path, &nd->last,
1397 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001398 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001399 } while (res > 0);
1400
1401 current->link_count--;
1402 nd->depth--;
1403 return res;
1404}
1405
1406/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001408 * This is the basic name resolution function, turning a pathname into
1409 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001411 * Returns 0 and nd will have valid dentry and mnt on success.
1412 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 */
Al Viro6de88d72009-08-09 01:41:57 +04001414static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415{
1416 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 int err;
1418 unsigned int lookup_flags = nd->flags;
1419
1420 while (*name=='/')
1421 name++;
1422 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001423 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 /* At this point we know we have a real path component. */
1426 for(;;) {
1427 unsigned long hash;
1428 struct qstr this;
1429 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001430 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Trond Myklebustcdce5d62005-10-18 14:20:18 -07001432 nd->flags |= LOOKUP_CONTINUE;
Al Viro52094c82011-02-21 21:34:47 -05001433
1434 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 if (err)
1436 break;
1437
1438 this.name = name;
1439 c = *(const unsigned char *)name;
1440
1441 hash = init_name_hash();
1442 do {
1443 name++;
1444 hash = partial_name_hash(c, hash);
1445 c = *(const unsigned char *)name;
1446 } while (c && (c != '/'));
1447 this.len = name - (const char *) this.name;
1448 this.hash = end_name_hash(hash);
1449
Al Virofe479a52011-02-22 15:10:03 -05001450 type = LAST_NORM;
1451 if (this.name[0] == '.') switch (this.len) {
1452 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001453 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001454 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001455 nd->flags |= LOOKUP_JUMPED;
1456 }
Al Virofe479a52011-02-22 15:10:03 -05001457 break;
1458 case 1:
1459 type = LAST_DOT;
1460 }
Al Viro5a202bc2011-03-08 14:17:44 -05001461 if (likely(type == LAST_NORM)) {
1462 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001463 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001464 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1465 err = parent->d_op->d_hash(parent, nd->inode,
1466 &this);
1467 if (err < 0)
1468 break;
1469 }
1470 }
Al Virofe479a52011-02-22 15:10:03 -05001471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 /* remove trailing slashes? */
1473 if (!c)
1474 goto last_component;
1475 while (*++name == '/');
1476 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001477 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
Al Viroce57dfc2011-03-13 19:58:58 -04001479 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1480 if (err < 0)
1481 return err;
Al Virofe479a52011-02-22 15:10:03 -05001482
Al Viroce57dfc2011-03-13 19:58:58 -04001483 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001484 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001486 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001489 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 break;
1491 continue;
1492 /* here ends the main loop */
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -08001495 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1496 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Al Virob3563792011-03-14 21:54:55 -04001497 nd->last = this;
1498 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001499 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 }
Al Viro951361f2011-03-04 14:44:37 -05001501 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 return err;
1503}
1504
Al Viro70e9b352011-03-05 21:12:22 -05001505static int path_init(int dfd, const char *name, unsigned int flags,
1506 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001508 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001509 int fput_needed;
1510 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001513 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001515 if (flags & LOOKUP_ROOT) {
1516 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001517 if (*name) {
1518 if (!inode->i_op->lookup)
1519 return -ENOTDIR;
1520 retval = inode_permission(inode, MAY_EXEC);
1521 if (retval)
1522 return retval;
1523 }
Al Viro5b6ca022011-03-09 23:04:47 -05001524 nd->path = nd->root;
1525 nd->inode = inode;
1526 if (flags & LOOKUP_RCU) {
1527 br_read_lock(vfsmount_lock);
1528 rcu_read_lock();
1529 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1530 } else {
1531 path_get(&nd->path);
1532 }
1533 return 0;
1534 }
1535
Al Viro2a737872009-04-07 11:49:53 -04001536 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001539 if (flags & LOOKUP_RCU) {
1540 br_read_lock(vfsmount_lock);
1541 rcu_read_lock();
1542 set_root_rcu(nd);
1543 } else {
1544 set_root(nd);
1545 path_get(&nd->root);
1546 }
Al Viro2a737872009-04-07 11:49:53 -04001547 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001548 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001549 if (flags & LOOKUP_RCU) {
1550 struct fs_struct *fs = current->fs;
1551 unsigned seq;
1552
1553 br_read_lock(vfsmount_lock);
1554 rcu_read_lock();
1555
1556 do {
1557 seq = read_seqcount_begin(&fs->seq);
1558 nd->path = fs->pwd;
1559 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1560 } while (read_seqcount_retry(&fs->seq, seq));
1561 } else {
1562 get_fs_pwd(current->fs, &nd->path);
1563 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001564 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001565 struct dentry *dentry;
1566
Al Viro1abf0c72011-03-13 03:51:11 -04001567 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001568 retval = -EBADF;
1569 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001570 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001571
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001572 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001573
Al Virof52e0c12011-03-14 18:56:51 -04001574 if (*name) {
1575 retval = -ENOTDIR;
1576 if (!S_ISDIR(dentry->d_inode->i_mode))
1577 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001578
Al Virof52e0c12011-03-14 18:56:51 -04001579 retval = file_permission(file, MAY_EXEC);
1580 if (retval)
1581 goto fput_fail;
1582 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001583
Jan Blunck5dd784d02008-02-14 19:34:38 -08001584 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001585 if (flags & LOOKUP_RCU) {
1586 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001587 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001588 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1589 br_read_lock(vfsmount_lock);
1590 rcu_read_lock();
1591 } else {
1592 path_get(&file->f_path);
1593 fput_light(file, fput_needed);
1594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 }
Al Viroe41f7d42011-02-22 14:02:58 -05001596
Nick Piggin31e6b012011-01-07 17:49:52 +11001597 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001598 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001599
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001600fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001601 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001602out_fail:
1603 return retval;
1604}
1605
Al Virobd92d7f2011-03-14 19:54:59 -04001606static inline int lookup_last(struct nameidata *nd, struct path *path)
1607{
1608 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1609 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1610
1611 nd->flags &= ~LOOKUP_PARENT;
1612 return walk_component(nd, path, &nd->last, nd->last_type,
1613 nd->flags & LOOKUP_FOLLOW);
1614}
1615
Al Viro9b4a9b12009-04-07 11:44:16 -04001616/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001617static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001618 unsigned int flags, struct nameidata *nd)
1619{
Al Viro70e9b352011-03-05 21:12:22 -05001620 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001621 struct path path;
1622 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001623
1624 /*
1625 * Path walking is largely split up into 2 different synchronisation
1626 * schemes, rcu-walk and ref-walk (explained in
1627 * Documentation/filesystems/path-lookup.txt). These share much of the
1628 * path walk code, but some things particularly setup, cleanup, and
1629 * following mounts are sufficiently divergent that functions are
1630 * duplicated. Typically there is a function foo(), and its RCU
1631 * analogue, foo_rcu().
1632 *
1633 * -ECHILD is the error number of choice (just to avoid clashes) that
1634 * is returned if some aspect of an rcu-walk fails. Such an error must
1635 * be handled by restarting a traditional ref-walk (which will always
1636 * be able to complete).
1637 */
Al Virobd92d7f2011-03-14 19:54:59 -04001638 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001639
Al Virobd92d7f2011-03-14 19:54:59 -04001640 if (unlikely(err))
1641 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001642
1643 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001644 err = link_path_walk(name, nd);
1645
1646 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001647 err = lookup_last(nd, &path);
1648 while (err > 0) {
1649 void *cookie;
1650 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001651 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001652 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001653 if (!err)
1654 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001655 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001656 }
1657 }
Al Viroee0827c2011-02-21 23:38:09 -05001658
1659 if (nd->flags & LOOKUP_RCU) {
Al Viro4455ca62011-03-04 14:28:10 -05001660 /* went all way through without dropping RCU */
Al Virobd92d7f2011-03-14 19:54:59 -04001661 BUG_ON(err);
Al Viro4455ca62011-03-04 14:28:10 -05001662 if (nameidata_drop_rcu_last(nd))
Al Virobd92d7f2011-03-14 19:54:59 -04001663 err = -ECHILD;
Al Viroee0827c2011-02-21 23:38:09 -05001664 }
1665
Al Virobd23a532011-03-23 09:56:30 -04001666 if (!err) {
Al Virobd92d7f2011-03-14 19:54:59 -04001667 err = handle_reval_path(nd);
Al Virobd23a532011-03-23 09:56:30 -04001668 if (err)
1669 path_put(&nd->path);
1670 }
Al Virobd92d7f2011-03-14 19:54:59 -04001671
1672 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1673 if (!nd->inode->i_op->lookup) {
1674 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001675 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001676 }
1677 }
Al Viro16c2cd72011-02-22 15:50:10 -05001678
Al Viro70e9b352011-03-05 21:12:22 -05001679 if (base)
1680 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001681
Al Viro5b6ca022011-03-09 23:04:47 -05001682 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001683 path_put(&nd->root);
1684 nd->root.mnt = NULL;
1685 }
Al Virobd92d7f2011-03-14 19:54:59 -04001686 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001687}
Nick Piggin31e6b012011-01-07 17:49:52 +11001688
Al Viroee0827c2011-02-21 23:38:09 -05001689static int do_path_lookup(int dfd, const char *name,
1690 unsigned int flags, struct nameidata *nd)
1691{
1692 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1693 if (unlikely(retval == -ECHILD))
1694 retval = path_lookupat(dfd, name, flags, nd);
1695 if (unlikely(retval == -ESTALE))
1696 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001697
1698 if (likely(!retval)) {
1699 if (unlikely(!audit_dummy_context())) {
1700 if (nd->path.dentry && nd->inode)
1701 audit_inode(name, nd->path.dentry);
1702 }
1703 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001704 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705}
1706
Al Viroc9c6cac2011-02-16 15:15:47 -05001707int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001708{
Al Viroc9c6cac2011-02-16 15:15:47 -05001709 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001710}
1711
Al Virod1811462008-08-02 00:49:18 -04001712int kern_path(const char *name, unsigned int flags, struct path *path)
1713{
1714 struct nameidata nd;
1715 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1716 if (!res)
1717 *path = nd.path;
1718 return res;
1719}
1720
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001721/**
1722 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1723 * @dentry: pointer to dentry of the base directory
1724 * @mnt: pointer to vfs mount of the base directory
1725 * @name: pointer to file name
1726 * @flags: lookup flags
1727 * @nd: pointer to nameidata
1728 */
1729int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1730 const char *name, unsigned int flags,
1731 struct nameidata *nd)
1732{
Al Viro5b6ca022011-03-09 23:04:47 -05001733 nd->root.dentry = dentry;
1734 nd->root.mnt = mnt;
1735 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1736 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001737}
1738
Christoph Hellwigeead1912007-10-16 23:25:38 -07001739static struct dentry *__lookup_hash(struct qstr *name,
1740 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001742 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001743 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 int err;
1745
Nick Pigginb74c79e2011-01-07 17:49:58 +11001746 err = exec_permission(inode, 0);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001747 if (err)
1748 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
1750 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001751 * Don't bother with __d_lookup: callers are for creat as
1752 * well as unlink, so a lot of the time it would cost
1753 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001754 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001755 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001756
Nick Pigginfb045ad2011-01-07 17:49:55 +11001757 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro6e6b1bd2009-08-13 23:38:37 +04001758 dentry = do_revalidate(dentry, nd);
1759
Nick Pigginbaa03892010-08-18 04:37:31 +10001760 if (!dentry)
1761 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 return dentry;
1764}
1765
James Morris057f6c02007-04-26 00:12:05 -07001766/*
1767 * Restricted form of lookup. Doesn't follow links, single-component only,
1768 * needs parent already locked. Doesn't follow mounts.
1769 * SMP-safe.
1770 */
Adrian Bunka244e162006-03-31 02:32:11 -08001771static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772{
Jan Blunck4ac91372008-02-14 19:34:32 -08001773 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774}
1775
Christoph Hellwigeead1912007-10-16 23:25:38 -07001776/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001777 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001778 * @name: pathname component to lookup
1779 * @base: base directory to lookup from
1780 * @len: maximum length @len should be interpreted to
1781 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001782 * Note that this routine is purely a helper for filesystem usage and should
1783 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001784 * nameidata argument is passed to the filesystem methods and a filesystem
1785 * using this helper needs to be prepared for that.
1786 */
James Morris057f6c02007-04-26 00:12:05 -07001787struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1788{
James Morris057f6c02007-04-26 00:12:05 -07001789 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001790 unsigned long hash;
1791 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001792
David Woodhouse2f9092e2009-04-20 23:18:37 +01001793 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1794
Al Viro6a96ba52011-03-07 23:49:20 -05001795 this.name = name;
1796 this.len = len;
1797 if (!len)
1798 return ERR_PTR(-EACCES);
1799
1800 hash = init_name_hash();
1801 while (len--) {
1802 c = *(const unsigned char *)name++;
1803 if (c == '/' || c == '\0')
1804 return ERR_PTR(-EACCES);
1805 hash = partial_name_hash(c, hash);
1806 }
1807 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001808 /*
1809 * See if the low-level filesystem might want
1810 * to use its own hash..
1811 */
1812 if (base->d_flags & DCACHE_OP_HASH) {
1813 int err = base->d_op->d_hash(base, base->d_inode, &this);
1814 if (err < 0)
1815 return ERR_PTR(err);
1816 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001817
Christoph Hellwig49705b72005-11-08 21:35:06 -08001818 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001819}
1820
Al Viro2d8f3032008-07-22 09:59:21 -04001821int user_path_at(int dfd, const char __user *name, unsigned flags,
1822 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823{
Al Viro2d8f3032008-07-22 09:59:21 -04001824 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001825 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001828
1829 BUG_ON(flags & LOOKUP_PARENT);
1830
1831 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001833 if (!err)
1834 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 }
1836 return err;
1837}
1838
Al Viro2ad94ae2008-07-21 09:32:51 -04001839static int user_path_parent(int dfd, const char __user *path,
1840 struct nameidata *nd, char **name)
1841{
1842 char *s = getname(path);
1843 int error;
1844
1845 if (IS_ERR(s))
1846 return PTR_ERR(s);
1847
1848 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1849 if (error)
1850 putname(s);
1851 else
1852 *name = s;
1853
1854 return error;
1855}
1856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857/*
1858 * It's inline, so penalty for filesystems that don't use sticky bit is
1859 * minimal.
1860 */
1861static inline int check_sticky(struct inode *dir, struct inode *inode)
1862{
David Howellsda9592e2008-11-14 10:39:05 +11001863 uid_t fsuid = current_fsuid();
1864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 if (!(dir->i_mode & S_ISVTX))
1866 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001867 if (current_user_ns() != inode_userns(inode))
1868 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001869 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001871 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001873
1874other_userns:
1875 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876}
1877
1878/*
1879 * Check whether we can remove a link victim from directory dir, check
1880 * whether the type of victim is right.
1881 * 1. We can't do it if dir is read-only (done in permission())
1882 * 2. We should have write and exec permissions on dir
1883 * 3. We can't remove anything from append-only dir
1884 * 4. We can't do anything with immutable dir (done in permission())
1885 * 5. If the sticky bit on dir is set we should either
1886 * a. be owner of dir, or
1887 * b. be owner of victim, or
1888 * c. have CAP_FOWNER capability
1889 * 6. If the victim is append-only or immutable we can't do antyhing with
1890 * links pointing to it.
1891 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1892 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1893 * 9. We can't remove a root or mountpoint.
1894 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1895 * nfs_async_unlink().
1896 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001897static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
1899 int error;
1900
1901 if (!victim->d_inode)
1902 return -ENOENT;
1903
1904 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001905 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
Al Virof419a2e2008-07-22 00:07:17 -04001907 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 if (error)
1909 return error;
1910 if (IS_APPEND(dir))
1911 return -EPERM;
1912 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001913 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 return -EPERM;
1915 if (isdir) {
1916 if (!S_ISDIR(victim->d_inode->i_mode))
1917 return -ENOTDIR;
1918 if (IS_ROOT(victim))
1919 return -EBUSY;
1920 } else if (S_ISDIR(victim->d_inode->i_mode))
1921 return -EISDIR;
1922 if (IS_DEADDIR(dir))
1923 return -ENOENT;
1924 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1925 return -EBUSY;
1926 return 0;
1927}
1928
1929/* Check whether we can create an object with dentry child in directory
1930 * dir.
1931 * 1. We can't do it if child already exists (open has special treatment for
1932 * this case, but since we are inlined it's OK)
1933 * 2. We can't do it if dir is read-only (done in permission())
1934 * 3. We should have write and exec permissions on dir
1935 * 4. We can't do it if dir is immutable (done in permission())
1936 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001937static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938{
1939 if (child->d_inode)
1940 return -EEXIST;
1941 if (IS_DEADDIR(dir))
1942 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001943 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944}
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946/*
1947 * p1 and p2 should be directories on the same fs.
1948 */
1949struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1950{
1951 struct dentry *p;
1952
1953 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001954 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return NULL;
1956 }
1957
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001958 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001960 p = d_ancestor(p2, p1);
1961 if (p) {
1962 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1963 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1964 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 }
1966
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001967 p = d_ancestor(p1, p2);
1968 if (p) {
1969 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1970 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1971 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 }
1973
Ingo Molnarf2eace22006-07-03 00:25:05 -07001974 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1975 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 return NULL;
1977}
1978
1979void unlock_rename(struct dentry *p1, struct dentry *p2)
1980{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001981 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001983 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001984 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 }
1986}
1987
1988int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1989 struct nameidata *nd)
1990{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001991 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
1993 if (error)
1994 return error;
1995
Al Viroacfa4382008-12-04 10:06:33 -05001996 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return -EACCES; /* shouldn't it be ENOSYS? */
1998 mode &= S_IALLUGO;
1999 mode |= S_IFREG;
2000 error = security_inode_create(dir, dentry, mode);
2001 if (error)
2002 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002004 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002005 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 return error;
2007}
2008
Al Viro73d049a2011-03-11 12:08:24 -05002009static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002011 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 struct inode *inode = dentry->d_inode;
2013 int error;
2014
Al Virobcda7652011-03-13 16:42:14 -04002015 /* O_PATH? */
2016 if (!acc_mode)
2017 return 0;
2018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 if (!inode)
2020 return -ENOENT;
2021
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002022 switch (inode->i_mode & S_IFMT) {
2023 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002025 case S_IFDIR:
2026 if (acc_mode & MAY_WRITE)
2027 return -EISDIR;
2028 break;
2029 case S_IFBLK:
2030 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002031 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002033 /*FALLTHRU*/
2034 case S_IFIFO:
2035 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002037 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002038 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002039
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002040 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002041 if (error)
2042 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 /*
2045 * An append-only file must be opened in append mode for writing.
2046 */
2047 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002048 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002049 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002051 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 }
2053
2054 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002055 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002056 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
2058 /*
2059 * Ensure there are no outstanding leases on the file.
2060 */
Al Virob65a9cf2009-12-16 06:27:40 -05002061 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002062}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
Jeff Laytone1181ee2010-12-07 16:19:50 -05002064static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002065{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002066 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002067 struct inode *inode = path->dentry->d_inode;
2068 int error = get_write_access(inode);
2069 if (error)
2070 return error;
2071 /*
2072 * Refuse to truncate files with mandatory locks held on them.
2073 */
2074 error = locks_verify_locked(inode);
2075 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002076 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002077 if (!error) {
2078 error = do_truncate(path->dentry, 0,
2079 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002080 filp);
Al Viro7715b522009-12-16 03:54:00 -05002081 }
2082 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002083 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084}
2085
Dave Hansend57999e2008-02-15 14:37:27 -08002086/*
Dave Hansend57999e2008-02-15 14:37:27 -08002087 * Note that while the flag value (low two bits) for sys_open means:
2088 * 00 - read-only
2089 * 01 - write-only
2090 * 10 - read-write
2091 * 11 - special
2092 * it is changed into
2093 * 00 - no permissions needed
2094 * 01 - read-permission
2095 * 10 - write-permission
2096 * 11 - read-write
2097 * for the internal routines (ie open_namei()/follow_link() etc)
2098 * This is more logical, and also allows the 00 "no perm needed"
2099 * to be used for symlinks (where the permissions are checked
2100 * later).
2101 *
2102*/
2103static inline int open_to_namei_flags(int flag)
2104{
2105 if ((flag+1) & O_ACCMODE)
2106 flag++;
2107 return flag;
2108}
2109
Nick Piggin31e6b012011-01-07 17:49:52 +11002110/*
Al Virofe2d35f2011-03-05 22:58:25 -05002111 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002112 */
Al Virofb1cc552009-12-24 01:58:28 -05002113static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002114 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002115{
Al Viroa1e28032009-12-24 02:12:06 -05002116 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002117 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002118 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002119 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002120 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002121 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002122 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002123 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002124
Al Viroc3e380b2011-02-23 13:39:45 -05002125 nd->flags &= ~LOOKUP_PARENT;
2126 nd->flags |= op->intent;
2127
Al Viro1f36f772009-12-26 10:56:19 -05002128 switch (nd->last_type) {
2129 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002130 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002131 error = handle_dots(nd, nd->last_type);
2132 if (error)
2133 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002134 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002135 case LAST_ROOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002136 if (nd->flags & LOOKUP_RCU) {
2137 if (nameidata_drop_rcu_last(nd))
2138 return ERR_PTR(-ECHILD);
2139 }
Al Viro16c2cd72011-02-22 15:50:10 -05002140 error = handle_reval_path(nd);
2141 if (error)
2142 goto exit;
Al Virofe2d35f2011-03-05 22:58:25 -05002143 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002144 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002145 error = -EISDIR;
2146 goto exit;
2147 }
2148 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002149 case LAST_BIND:
Al Virofe2d35f2011-03-05 22:58:25 -05002150 /* can't be RCU mode here */
Al Viro16c2cd72011-02-22 15:50:10 -05002151 error = handle_reval_path(nd);
2152 if (error)
2153 goto exit;
Al Viro1f36f772009-12-26 10:56:19 -05002154 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002155 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002156 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002157
Al Viroca344a892011-03-09 00:36:45 -05002158 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002159 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002160 if (nd->last.name[nd->last.len])
2161 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002162 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2163 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002164 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002165 error = walk_component(nd, path, &nd->last, LAST_NORM,
2166 !symlink_ok);
2167 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002168 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002169 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002170 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002171 /* sayonara */
2172 if (nd->flags & LOOKUP_RCU) {
2173 if (nameidata_drop_rcu_last(nd))
2174 return ERR_PTR(-ECHILD);
2175 }
2176
2177 error = -ENOTDIR;
2178 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002179 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002180 goto exit;
2181 }
2182 audit_inode(pathname, nd->path.dentry);
2183 goto ok;
2184 }
2185
2186 /* create side of things */
2187
2188 if (nd->flags & LOOKUP_RCU) {
2189 if (nameidata_drop_rcu_last(nd))
2190 return ERR_PTR(-ECHILD);
2191 }
2192
2193 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002194 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002195 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002196 if (nd->last.name[nd->last.len])
2197 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002198
Al Viroa1e28032009-12-24 02:12:06 -05002199 mutex_lock(&dir->d_inode->i_mutex);
2200
Al Viro6c0d46c2011-03-09 00:59:59 -05002201 dentry = lookup_hash(nd);
2202 error = PTR_ERR(dentry);
2203 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002204 mutex_unlock(&dir->d_inode->i_mutex);
2205 goto exit;
2206 }
2207
Al Viro6c0d46c2011-03-09 00:59:59 -05002208 path->dentry = dentry;
2209 path->mnt = nd->path.mnt;
2210
Al Virofb1cc552009-12-24 01:58:28 -05002211 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002212 if (!dentry->d_inode) {
2213 int mode = op->mode;
2214 if (!IS_POSIXACL(dir->d_inode))
2215 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002216 /*
2217 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002218 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002219 * the time when the file is created and when
2220 * a permanent write count is taken through
2221 * the 'struct file' in nameidata_to_filp().
2222 */
2223 error = mnt_want_write(nd->path.mnt);
2224 if (error)
2225 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002226 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002227 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002228 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002229 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002230 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002231 error = security_path_mknod(&nd->path, dentry, mode, 0);
2232 if (error)
2233 goto exit_mutex_unlock;
2234 error = vfs_create(dir->d_inode, dentry, mode, nd);
2235 if (error)
2236 goto exit_mutex_unlock;
2237 mutex_unlock(&dir->d_inode->i_mutex);
2238 dput(nd->path.dentry);
2239 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002240 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002241 }
2242
2243 /*
2244 * It already exists.
2245 */
2246 mutex_unlock(&dir->d_inode->i_mutex);
2247 audit_inode(pathname, path->dentry);
2248
2249 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002250 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002251 goto exit_dput;
2252
David Howells9875cf82011-01-14 18:45:21 +00002253 error = follow_managed(path, nd->flags);
2254 if (error < 0)
2255 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002256
2257 error = -ENOENT;
2258 if (!path->dentry->d_inode)
2259 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002260
2261 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002262 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002263
2264 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002265 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002266 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002267 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002268 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002269ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002270 if (!S_ISREG(nd->inode->i_mode))
2271 will_truncate = 0;
2272
Al Viro0f9d1a12011-03-09 00:13:14 -05002273 if (will_truncate) {
2274 error = mnt_want_write(nd->path.mnt);
2275 if (error)
2276 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002277 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002278 }
Al Viroca344a892011-03-09 00:36:45 -05002279common:
Al Virobcda7652011-03-13 16:42:14 -04002280 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002281 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002282 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002283 filp = nameidata_to_filp(nd);
2284 if (!IS_ERR(filp)) {
2285 error = ima_file_check(filp, op->acc_mode);
2286 if (error) {
2287 fput(filp);
2288 filp = ERR_PTR(error);
2289 }
2290 }
2291 if (!IS_ERR(filp)) {
2292 if (will_truncate) {
2293 error = handle_truncate(filp);
2294 if (error) {
2295 fput(filp);
2296 filp = ERR_PTR(error);
2297 }
2298 }
2299 }
Al Viroca344a892011-03-09 00:36:45 -05002300out:
2301 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002302 mnt_drop_write(nd->path.mnt);
2303 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002304 return filp;
2305
2306exit_mutex_unlock:
2307 mutex_unlock(&dir->d_inode->i_mutex);
2308exit_dput:
2309 path_put_conditional(path, nd);
2310exit:
Al Viroca344a892011-03-09 00:36:45 -05002311 filp = ERR_PTR(error);
2312 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002313}
2314
Al Viro13aab422011-02-23 17:54:08 -05002315static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002316 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317{
Al Virofe2d35f2011-03-05 22:58:25 -05002318 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002319 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002320 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002321 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002322
2323 filp = get_empty_filp();
2324 if (!filp)
2325 return ERR_PTR(-ENFILE);
2326
Al Viro47c805d2011-02-23 17:44:09 -05002327 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002328 nd->intent.open.file = filp;
2329 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2330 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002331
Al Viro73d049a2011-03-11 12:08:24 -05002332 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002333 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002334 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002335
Al Virofe2d35f2011-03-05 22:58:25 -05002336 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002337 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002338 if (unlikely(error))
2339 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
Al Viro73d049a2011-03-11 12:08:24 -05002341 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002342 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002343 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002344 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002345 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002346 path_put_conditional(&path, nd);
2347 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002348 filp = ERR_PTR(-ELOOP);
2349 break;
2350 }
Al Viro73d049a2011-03-11 12:08:24 -05002351 nd->flags |= LOOKUP_PARENT;
2352 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002353 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002354 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002355 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002356 else
Al Viro73d049a2011-03-11 12:08:24 -05002357 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002358 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002359 }
Al Viro10fa8e62009-12-26 07:09:49 -05002360out:
Al Viro73d049a2011-03-11 12:08:24 -05002361 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2362 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002363 if (base)
2364 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002365 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002366 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
Nick Piggin31e6b012011-01-07 17:49:52 +11002368out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002369 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002370 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371}
2372
Al Viro13aab422011-02-23 17:54:08 -05002373struct file *do_filp_open(int dfd, const char *pathname,
2374 const struct open_flags *op, int flags)
2375{
Al Viro73d049a2011-03-11 12:08:24 -05002376 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002377 struct file *filp;
2378
Al Viro73d049a2011-03-11 12:08:24 -05002379 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002380 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002381 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002382 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002383 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002384 return filp;
2385}
2386
Al Viro73d049a2011-03-11 12:08:24 -05002387struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2388 const char *name, const struct open_flags *op, int flags)
2389{
2390 struct nameidata nd;
2391 struct file *file;
2392
2393 nd.root.mnt = mnt;
2394 nd.root.dentry = dentry;
2395
2396 flags |= LOOKUP_ROOT;
2397
Al Virobcda7652011-03-13 16:42:14 -04002398 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002399 return ERR_PTR(-ELOOP);
2400
2401 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2402 if (unlikely(file == ERR_PTR(-ECHILD)))
2403 file = path_openat(-1, name, &nd, op, flags);
2404 if (unlikely(file == ERR_PTR(-ESTALE)))
2405 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2406 return file;
2407}
2408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409/**
2410 * lookup_create - lookup a dentry, creating it if it doesn't exist
2411 * @nd: nameidata info
2412 * @is_dir: directory flag
2413 *
2414 * Simple function to lookup and return a dentry and create it
2415 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002416 *
Jan Blunck4ac91372008-02-14 19:34:32 -08002417 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 */
2419struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2420{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002421 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Jan Blunck4ac91372008-02-14 19:34:32 -08002423 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002424 /*
2425 * Yucky last component or no last component at all?
2426 * (foo/., foo/.., /////)
2427 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 if (nd->last_type != LAST_NORM)
2429 goto fail;
2430 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04002431 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04002432 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002433
2434 /*
2435 * Do the final lookup.
2436 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08002437 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (IS_ERR(dentry))
2439 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002440
Al Viroe9baf6e2008-05-15 04:49:12 -04002441 if (dentry->d_inode)
2442 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002443 /*
2444 * Special case - lookup gave negative, but... we had foo/bar/
2445 * From the vfs_mknod() POV we just have a negative dentry -
2446 * all is fine. Let's be bastards - you had / on the end, you've
2447 * been asking for (non-existent) directory. -ENOENT for you.
2448 */
Al Viroe9baf6e2008-05-15 04:49:12 -04002449 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2450 dput(dentry);
2451 dentry = ERR_PTR(-ENOENT);
2452 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002454eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002456 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457fail:
2458 return dentry;
2459}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07002460EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
2462int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2463{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002464 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
2466 if (error)
2467 return error;
2468
Serge E. Hallyne795b712011-03-23 16:43:25 -07002469 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2470 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 return -EPERM;
2472
Al Viroacfa4382008-12-04 10:06:33 -05002473 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 return -EPERM;
2475
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002476 error = devcgroup_inode_mknod(mode, dev);
2477 if (error)
2478 return error;
2479
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 error = security_inode_mknod(dir, dentry, mode, dev);
2481 if (error)
2482 return error;
2483
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002485 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002486 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 return error;
2488}
2489
Dave Hansen463c3192008-02-15 14:37:57 -08002490static int may_mknod(mode_t mode)
2491{
2492 switch (mode & S_IFMT) {
2493 case S_IFREG:
2494 case S_IFCHR:
2495 case S_IFBLK:
2496 case S_IFIFO:
2497 case S_IFSOCK:
2498 case 0: /* zero mode translates to S_IFREG */
2499 return 0;
2500 case S_IFDIR:
2501 return -EPERM;
2502 default:
2503 return -EINVAL;
2504 }
2505}
2506
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002507SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2508 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
Al Viro2ad94ae2008-07-21 09:32:51 -04002510 int error;
2511 char *tmp;
2512 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 struct nameidata nd;
2514
2515 if (S_ISDIR(mode))
2516 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
Al Viro2ad94ae2008-07-21 09:32:51 -04002518 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002520 return error;
2521
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002523 if (IS_ERR(dentry)) {
2524 error = PTR_ERR(dentry);
2525 goto out_unlock;
2526 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002527 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002528 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002529 error = may_mknod(mode);
2530 if (error)
2531 goto out_dput;
2532 error = mnt_want_write(nd.path.mnt);
2533 if (error)
2534 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002535 error = security_path_mknod(&nd.path, dentry, mode, dev);
2536 if (error)
2537 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002538 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002540 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 break;
2542 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002543 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 new_decode_dev(dev));
2545 break;
2546 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002547 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002550out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002551 mnt_drop_write(nd.path.mnt);
2552out_dput:
2553 dput(dentry);
2554out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002555 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002556 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 putname(tmp);
2558
2559 return error;
2560}
2561
Heiko Carstens3480b252009-01-14 14:14:16 +01002562SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002563{
2564 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2565}
2566
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2568{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002569 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
2571 if (error)
2572 return error;
2573
Al Viroacfa4382008-12-04 10:06:33 -05002574 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 return -EPERM;
2576
2577 mode &= (S_IRWXUGO|S_ISVTX);
2578 error = security_inode_mkdir(dir, dentry, mode);
2579 if (error)
2580 return error;
2581
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002583 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002584 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 return error;
2586}
2587
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002588SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589{
2590 int error = 0;
2591 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002592 struct dentry *dentry;
2593 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
Al Viro2ad94ae2008-07-21 09:32:51 -04002595 error = user_path_parent(dfd, pathname, &nd, &tmp);
2596 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002597 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Dave Hansen6902d922006-09-30 23:29:01 -07002599 dentry = lookup_create(&nd, 1);
2600 error = PTR_ERR(dentry);
2601 if (IS_ERR(dentry))
2602 goto out_unlock;
2603
Jan Blunck4ac91372008-02-14 19:34:32 -08002604 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002605 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002606 error = mnt_want_write(nd.path.mnt);
2607 if (error)
2608 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002609 error = security_path_mkdir(&nd.path, dentry, mode);
2610 if (error)
2611 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002612 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002613out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002614 mnt_drop_write(nd.path.mnt);
2615out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002616 dput(dentry);
2617out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002618 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002619 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002620 putname(tmp);
2621out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 return error;
2623}
2624
Heiko Carstens3cdad422009-01-14 14:14:22 +01002625SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002626{
2627 return sys_mkdirat(AT_FDCWD, pathname, mode);
2628}
2629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630/*
2631 * We try to drop the dentry early: we should have
2632 * a usage count of 2 if we're the only user of this
2633 * dentry, and if that is true (possibly after pruning
2634 * the dcache), then we drop the dentry now.
2635 *
2636 * A low-level filesystem can, if it choses, legally
2637 * do a
2638 *
2639 * if (!d_unhashed(dentry))
2640 * return -EBUSY;
2641 *
2642 * if it cannot handle the case of removing a directory
2643 * that is still in use by something else..
2644 */
2645void dentry_unhash(struct dentry *dentry)
2646{
2647 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002648 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002650 if (dentry->d_count == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 __d_drop(dentry);
2652 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653}
2654
2655int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2656{
2657 int error = may_delete(dir, dentry, 1);
2658
2659 if (error)
2660 return error;
2661
Al Viroacfa4382008-12-04 10:06:33 -05002662 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 return -EPERM;
2664
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002665 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 dentry_unhash(dentry);
2667 if (d_mountpoint(dentry))
2668 error = -EBUSY;
2669 else {
2670 error = security_inode_rmdir(dir, dentry);
2671 if (!error) {
2672 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002673 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002675 dont_mount(dentry);
2676 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 }
2678 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002679 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 d_delete(dentry);
2682 }
2683 dput(dentry);
2684
2685 return error;
2686}
2687
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002688static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689{
2690 int error = 0;
2691 char * name;
2692 struct dentry *dentry;
2693 struct nameidata nd;
2694
Al Viro2ad94ae2008-07-21 09:32:51 -04002695 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002697 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698
2699 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002700 case LAST_DOTDOT:
2701 error = -ENOTEMPTY;
2702 goto exit1;
2703 case LAST_DOT:
2704 error = -EINVAL;
2705 goto exit1;
2706 case LAST_ROOT:
2707 error = -EBUSY;
2708 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002710
2711 nd.flags &= ~LOOKUP_PARENT;
2712
Jan Blunck4ac91372008-02-14 19:34:32 -08002713 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002714 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002716 if (IS_ERR(dentry))
2717 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002718 error = mnt_want_write(nd.path.mnt);
2719 if (error)
2720 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002721 error = security_path_rmdir(&nd.path, dentry);
2722 if (error)
2723 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002724 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002725exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002726 mnt_drop_write(nd.path.mnt);
2727exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002728 dput(dentry);
2729exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002730 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002732 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 putname(name);
2734 return error;
2735}
2736
Heiko Carstens3cdad422009-01-14 14:14:22 +01002737SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002738{
2739 return do_rmdir(AT_FDCWD, pathname);
2740}
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742int vfs_unlink(struct inode *dir, struct dentry *dentry)
2743{
2744 int error = may_delete(dir, dentry, 0);
2745
2746 if (error)
2747 return error;
2748
Al Viroacfa4382008-12-04 10:06:33 -05002749 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 return -EPERM;
2751
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002752 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 if (d_mountpoint(dentry))
2754 error = -EBUSY;
2755 else {
2756 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002757 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002759 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002760 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002763 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
2765 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2766 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002767 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 }
Robert Love0eeca282005-07-12 17:06:03 -04002770
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 return error;
2772}
2773
2774/*
2775 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002776 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 * writeout happening, and we don't want to prevent access to the directory
2778 * while waiting on the I/O.
2779 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002780static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781{
Al Viro2ad94ae2008-07-21 09:32:51 -04002782 int error;
2783 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 struct dentry *dentry;
2785 struct nameidata nd;
2786 struct inode *inode = NULL;
2787
Al Viro2ad94ae2008-07-21 09:32:51 -04002788 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002790 return error;
2791
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 error = -EISDIR;
2793 if (nd.last_type != LAST_NORM)
2794 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002795
2796 nd.flags &= ~LOOKUP_PARENT;
2797
Jan Blunck4ac91372008-02-14 19:34:32 -08002798 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002799 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 error = PTR_ERR(dentry);
2801 if (!IS_ERR(dentry)) {
2802 /* Why not before? Because we want correct error value */
2803 if (nd.last.name[nd.last.len])
2804 goto slashes;
2805 inode = dentry->d_inode;
2806 if (inode)
Al Viro7de9c6ee2010-10-23 11:11:40 -04002807 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002808 error = mnt_want_write(nd.path.mnt);
2809 if (error)
2810 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002811 error = security_path_unlink(&nd.path, dentry);
2812 if (error)
2813 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002814 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002815exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002816 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 exit2:
2818 dput(dentry);
2819 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002820 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 if (inode)
2822 iput(inode); /* truncate the inode here */
2823exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002824 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 putname(name);
2826 return error;
2827
2828slashes:
2829 error = !dentry->d_inode ? -ENOENT :
2830 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2831 goto exit2;
2832}
2833
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002834SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002835{
2836 if ((flag & ~AT_REMOVEDIR) != 0)
2837 return -EINVAL;
2838
2839 if (flag & AT_REMOVEDIR)
2840 return do_rmdir(dfd, pathname);
2841
2842 return do_unlinkat(dfd, pathname);
2843}
2844
Heiko Carstens3480b252009-01-14 14:14:16 +01002845SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002846{
2847 return do_unlinkat(AT_FDCWD, pathname);
2848}
2849
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002850int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002852 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853
2854 if (error)
2855 return error;
2856
Al Viroacfa4382008-12-04 10:06:33 -05002857 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 return -EPERM;
2859
2860 error = security_inode_symlink(dir, dentry, oldname);
2861 if (error)
2862 return error;
2863
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002865 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002866 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 return error;
2868}
2869
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002870SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2871 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872{
Al Viro2ad94ae2008-07-21 09:32:51 -04002873 int error;
2874 char *from;
2875 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002876 struct dentry *dentry;
2877 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878
2879 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002880 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002882
2883 error = user_path_parent(newdfd, newname, &nd, &to);
2884 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002885 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886
Dave Hansen6902d922006-09-30 23:29:01 -07002887 dentry = lookup_create(&nd, 0);
2888 error = PTR_ERR(dentry);
2889 if (IS_ERR(dentry))
2890 goto out_unlock;
2891
Dave Hansen75c3f292008-02-15 14:37:45 -08002892 error = mnt_want_write(nd.path.mnt);
2893 if (error)
2894 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002895 error = security_path_symlink(&nd.path, dentry, from);
2896 if (error)
2897 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002898 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002899out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002900 mnt_drop_write(nd.path.mnt);
2901out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002902 dput(dentry);
2903out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002904 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002905 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002906 putname(to);
2907out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 putname(from);
2909 return error;
2910}
2911
Heiko Carstens3480b252009-01-14 14:14:16 +01002912SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002913{
2914 return sys_symlinkat(oldname, AT_FDCWD, newname);
2915}
2916
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2918{
2919 struct inode *inode = old_dentry->d_inode;
2920 int error;
2921
2922 if (!inode)
2923 return -ENOENT;
2924
Miklos Szeredia95164d2008-07-30 15:08:48 +02002925 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 if (error)
2927 return error;
2928
2929 if (dir->i_sb != inode->i_sb)
2930 return -EXDEV;
2931
2932 /*
2933 * A link to an append-only or immutable file cannot be created.
2934 */
2935 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2936 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002937 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002939 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 return -EPERM;
2941
2942 error = security_inode_link(old_dentry, dir, new_dentry);
2943 if (error)
2944 return error;
2945
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002946 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302947 /* Make sure we don't allow creating hardlink to an unlinked file */
2948 if (inode->i_nlink == 0)
2949 error = -ENOENT;
2950 else
2951 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002952 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002953 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002954 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 return error;
2956}
2957
2958/*
2959 * Hardlinks are often used in delicate situations. We avoid
2960 * security-related surprises by not following symlinks on the
2961 * newname. --KAB
2962 *
2963 * We don't follow them on the oldname either to be compatible
2964 * with linux 2.0, and to avoid hard-linking to directories
2965 * and other special files. --ADM
2966 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002967SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2968 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969{
2970 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002971 struct nameidata nd;
2972 struct path old_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302973 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002975 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302977 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002978 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302979 /*
2980 * To use null names we require CAP_DAC_READ_SEARCH
2981 * This ensures that not everyone will be able to create
2982 * handlink using the passed filedescriptor.
2983 */
2984 if (flags & AT_EMPTY_PATH) {
2985 if (!capable(CAP_DAC_READ_SEARCH))
2986 return -ENOENT;
2987 how = LOOKUP_EMPTY;
2988 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002989
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302990 if (flags & AT_SYMLINK_FOLLOW)
2991 how |= LOOKUP_FOLLOW;
2992
2993 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002995 return error;
2996
2997 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 if (error)
2999 goto out;
3000 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04003001 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 goto out_release;
3003 new_dentry = lookup_create(&nd, 0);
3004 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003005 if (IS_ERR(new_dentry))
3006 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08003007 error = mnt_want_write(nd.path.mnt);
3008 if (error)
3009 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003010 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3011 if (error)
3012 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04003013 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003014out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08003015 mnt_drop_write(nd.path.mnt);
3016out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003017 dput(new_dentry);
3018out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08003019 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08003021 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003022 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023out:
Al Viro2d8f3032008-07-22 09:59:21 -04003024 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
3026 return error;
3027}
3028
Heiko Carstens3480b252009-01-14 14:14:16 +01003029SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003030{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003031 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003032}
3033
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034/*
3035 * The worst of all namespace operations - renaming directory. "Perverted"
3036 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3037 * Problems:
3038 * a) we can get into loop creation. Check is done in is_subdir().
3039 * b) race potential - two innocent renames can create a loop together.
3040 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003041 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 * story.
3043 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003044 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 * whether the target exists). Solution: try to be smart with locking
3046 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003047 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 * move will be locked. Thus we can rank directories by the tree
3049 * (ancestors first) and rank all non-directories after them.
3050 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003051 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 * HOWEVER, it relies on the assumption that any object with ->lookup()
3053 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3054 * we'd better make sure that there's no link(2) for them.
3055 * d) some filesystems don't support opened-but-unlinked directories,
3056 * either because of layout or because they are not ready to deal with
3057 * all cases correctly. The latter will be fixed (taking this sort of
3058 * stuff into VFS), but the former is not going away. Solution: the same
3059 * trick as in rmdir().
3060 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003061 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003063 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 * locking].
3065 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003066static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3067 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068{
3069 int error = 0;
3070 struct inode *target;
3071
3072 /*
3073 * If we are going to change the parent - check write permissions,
3074 * we'll need to flip '..'.
3075 */
3076 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003077 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078 if (error)
3079 return error;
3080 }
3081
3082 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3083 if (error)
3084 return error;
3085
3086 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04003087 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003088 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3090 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04003091 else {
3092 if (target)
3093 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04003097 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04003099 dont_mount(new_dentry);
3100 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003101 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 if (d_unhashed(new_dentry))
3103 d_rehash(new_dentry);
3104 dput(new_dentry);
3105 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003106 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003107 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3108 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 return error;
3110}
3111
Adrian Bunk75c96f82005-05-05 16:16:09 -07003112static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3113 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114{
3115 struct inode *target;
3116 int error;
3117
3118 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3119 if (error)
3120 return error;
3121
3122 dget(new_dentry);
3123 target = new_dentry->d_inode;
3124 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003125 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3127 error = -EBUSY;
3128 else
3129 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3130 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05003131 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04003132 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07003133 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 }
3136 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003137 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138 dput(new_dentry);
3139 return error;
3140}
3141
3142int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3143 struct inode *new_dir, struct dentry *new_dentry)
3144{
3145 int error;
3146 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003147 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148
3149 if (old_dentry->d_inode == new_dentry->d_inode)
3150 return 0;
3151
3152 error = may_delete(old_dir, old_dentry, is_dir);
3153 if (error)
3154 return error;
3155
3156 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003157 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 else
3159 error = may_delete(new_dir, new_dentry, is_dir);
3160 if (error)
3161 return error;
3162
Al Viroacfa4382008-12-04 10:06:33 -05003163 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 return -EPERM;
3165
Robert Love0eeca282005-07-12 17:06:03 -04003166 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3167
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 if (is_dir)
3169 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3170 else
3171 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003172 if (!error)
3173 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003174 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003175 fsnotify_oldname_free(old_name);
3176
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 return error;
3178}
3179
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003180SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3181 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182{
Al Viro2ad94ae2008-07-21 09:32:51 -04003183 struct dentry *old_dir, *new_dir;
3184 struct dentry *old_dentry, *new_dentry;
3185 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003187 char *from;
3188 char *to;
3189 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190
Al Viro2ad94ae2008-07-21 09:32:51 -04003191 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 if (error)
3193 goto exit;
3194
Al Viro2ad94ae2008-07-21 09:32:51 -04003195 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 if (error)
3197 goto exit1;
3198
3199 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003200 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201 goto exit2;
3202
Jan Blunck4ac91372008-02-14 19:34:32 -08003203 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 error = -EBUSY;
3205 if (oldnd.last_type != LAST_NORM)
3206 goto exit2;
3207
Jan Blunck4ac91372008-02-14 19:34:32 -08003208 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 if (newnd.last_type != LAST_NORM)
3210 goto exit2;
3211
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003212 oldnd.flags &= ~LOOKUP_PARENT;
3213 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003214 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003215
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 trap = lock_rename(new_dir, old_dir);
3217
Christoph Hellwig49705b72005-11-08 21:35:06 -08003218 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 error = PTR_ERR(old_dentry);
3220 if (IS_ERR(old_dentry))
3221 goto exit3;
3222 /* source must exist */
3223 error = -ENOENT;
3224 if (!old_dentry->d_inode)
3225 goto exit4;
3226 /* unless the source is a directory trailing slashes give -ENOTDIR */
3227 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3228 error = -ENOTDIR;
3229 if (oldnd.last.name[oldnd.last.len])
3230 goto exit4;
3231 if (newnd.last.name[newnd.last.len])
3232 goto exit4;
3233 }
3234 /* source should not be ancestor of target */
3235 error = -EINVAL;
3236 if (old_dentry == trap)
3237 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003238 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 error = PTR_ERR(new_dentry);
3240 if (IS_ERR(new_dentry))
3241 goto exit4;
3242 /* target should not be an ancestor of source */
3243 error = -ENOTEMPTY;
3244 if (new_dentry == trap)
3245 goto exit5;
3246
Dave Hansen9079b1e2008-02-15 14:37:49 -08003247 error = mnt_want_write(oldnd.path.mnt);
3248 if (error)
3249 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003250 error = security_path_rename(&oldnd.path, old_dentry,
3251 &newnd.path, new_dentry);
3252 if (error)
3253 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 error = vfs_rename(old_dir->d_inode, old_dentry,
3255 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003256exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003257 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258exit5:
3259 dput(new_dentry);
3260exit4:
3261 dput(old_dentry);
3262exit3:
3263 unlock_rename(new_dir, old_dir);
3264exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003265 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003266 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003268 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003270exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 return error;
3272}
3273
Heiko Carstensa26eab22009-01-14 14:14:17 +01003274SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003275{
3276 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3277}
3278
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3280{
3281 int len;
3282
3283 len = PTR_ERR(link);
3284 if (IS_ERR(link))
3285 goto out;
3286
3287 len = strlen(link);
3288 if (len > (unsigned) buflen)
3289 len = buflen;
3290 if (copy_to_user(buffer, link, len))
3291 len = -EFAULT;
3292out:
3293 return len;
3294}
3295
3296/*
3297 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3298 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3299 * using) it for any given inode is up to filesystem.
3300 */
3301int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3302{
3303 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003304 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003305 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003306
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003308 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003309 if (IS_ERR(cookie))
3310 return PTR_ERR(cookie);
3311
3312 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3313 if (dentry->d_inode->i_op->put_link)
3314 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3315 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316}
3317
3318int vfs_follow_link(struct nameidata *nd, const char *link)
3319{
3320 return __vfs_follow_link(nd, link);
3321}
3322
3323/* get the link contents into pagecache */
3324static char *page_getlink(struct dentry * dentry, struct page **ppage)
3325{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003326 char *kaddr;
3327 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003329 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003331 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003333 kaddr = kmap(page);
3334 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3335 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336}
3337
3338int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3339{
3340 struct page *page = NULL;
3341 char *s = page_getlink(dentry, &page);
3342 int res = vfs_readlink(dentry,buffer,buflen,s);
3343 if (page) {
3344 kunmap(page);
3345 page_cache_release(page);
3346 }
3347 return res;
3348}
3349
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003350void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003352 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003354 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355}
3356
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003357void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003359 struct page *page = cookie;
3360
3361 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 kunmap(page);
3363 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 }
3365}
3366
Nick Piggin54566b22009-01-04 12:00:53 -08003367/*
3368 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3369 */
3370int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371{
3372 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003373 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003374 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003375 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003377 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3378 if (nofs)
3379 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380
NeilBrown7e53cac2006-03-25 03:07:57 -08003381retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003382 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003383 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003385 goto fail;
3386
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 kaddr = kmap_atomic(page, KM_USER0);
3388 memcpy(kaddr, symname, len-1);
3389 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003390
3391 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3392 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 if (err < 0)
3394 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003395 if (err < len-1)
3396 goto retry;
3397
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 mark_inode_dirty(inode);
3399 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400fail:
3401 return err;
3402}
3403
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003404int page_symlink(struct inode *inode, const char *symname, int len)
3405{
3406 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003407 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003408}
3409
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003410const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 .readlink = generic_readlink,
3412 .follow_link = page_follow_link_light,
3413 .put_link = page_put_link,
3414};
3415
Al Viro2d8f3032008-07-22 09:59:21 -04003416EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003417EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418EXPORT_SYMBOL(follow_down);
3419EXPORT_SYMBOL(follow_up);
3420EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3421EXPORT_SYMBOL(getname);
3422EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423EXPORT_SYMBOL(lookup_one_len);
3424EXPORT_SYMBOL(page_follow_link_light);
3425EXPORT_SYMBOL(page_put_link);
3426EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003427EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428EXPORT_SYMBOL(page_symlink);
3429EXPORT_SYMBOL(page_symlink_inode_operations);
Al Viroc9c6cac2011-02-16 15:15:47 -05003430EXPORT_SYMBOL(kern_path_parent);
Al Virod1811462008-08-02 00:49:18 -04003431EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003432EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003433EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08003434EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435EXPORT_SYMBOL(unlock_rename);
3436EXPORT_SYMBOL(vfs_create);
3437EXPORT_SYMBOL(vfs_follow_link);
3438EXPORT_SYMBOL(vfs_link);
3439EXPORT_SYMBOL(vfs_mkdir);
3440EXPORT_SYMBOL(vfs_mknod);
3441EXPORT_SYMBOL(generic_permission);
3442EXPORT_SYMBOL(vfs_readlink);
3443EXPORT_SYMBOL(vfs_rename);
3444EXPORT_SYMBOL(vfs_rmdir);
3445EXPORT_SYMBOL(vfs_symlink);
3446EXPORT_SYMBOL(vfs_unlink);
3447EXPORT_SYMBOL(dentry_unhash);
3448EXPORT_SYMBOL(generic_readlink);