blob: 6ff858c049c030fd5ea1e142a02c39bb4b51a2dc [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
Lucas De Marchi25985ed2011-03-30 22:57:33 -030073 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 *
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{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700182 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700183
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;
Tim Chenc1530012011-04-15 11:39:29 -0700700 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100701 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100702 }
703}
704
Arjan van de Venf1662352006-01-14 13:21:31 -0800705static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
Nick Piggin31e6b012011-01-07 17:49:52 +1100707 int ret;
708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 if (IS_ERR(link))
710 goto fail;
711
712 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400713 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800714 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400715 nd->path = nd->root;
716 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500717 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100719 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100720
Nick Piggin31e6b012011-01-07 17:49:52 +1100721 ret = link_path_walk(link, nd);
722 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800724 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return PTR_ERR(link);
726}
727
Jan Blunck1d957f92008-02-14 19:34:35 -0800728static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700729{
730 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800731 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700732 mntput(path->mnt);
733}
734
Nick Piggin7b9337a2011-01-14 08:42:43 +0000735static inline void path_to_nameidata(const struct path *path,
736 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700737{
Nick Piggin31e6b012011-01-07 17:49:52 +1100738 if (!(nd->flags & LOOKUP_RCU)) {
739 dput(nd->path.dentry);
740 if (nd->path.mnt != path->mnt)
741 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800742 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100743 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800744 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700745}
746
Al Viro574197e2011-03-14 22:20:34 -0400747static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
748{
749 struct inode *inode = link->dentry->d_inode;
750 if (!IS_ERR(cookie) && inode->i_op->put_link)
751 inode->i_op->put_link(link->dentry, nd, cookie);
752 path_put(link);
753}
754
Al Virodef4af32009-12-26 08:37:05 -0500755static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400756follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800757{
758 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000759 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800760
Al Viro844a3912011-02-15 00:38:26 -0500761 BUG_ON(nd->flags & LOOKUP_RCU);
762
Al Viro0e794582011-03-16 02:45:02 -0400763 if (link->mnt == nd->path.mnt)
764 mntget(link->mnt);
765
Al Viro574197e2011-03-14 22:20:34 -0400766 if (unlikely(current->total_link_count >= 40)) {
767 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400768 path_put(&nd->path);
769 return -ELOOP;
770 }
771 cond_resched();
772 current->total_link_count++;
773
Nick Piggin7b9337a2011-01-14 08:42:43 +0000774 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800775 nd_set_link(nd, NULL);
776
Al Viro36f3b4f2011-02-22 21:24:38 -0500777 error = security_inode_follow_link(link->dentry, nd);
778 if (error) {
779 *p = ERR_PTR(error); /* no ->put_link(), please */
780 path_put(&nd->path);
781 return error;
782 }
783
Al Viro86acdca12009-12-22 23:45:11 -0500784 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500785 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
786 error = PTR_ERR(*p);
787 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800788 char *s = nd_get_link(nd);
789 error = 0;
790 if (s)
791 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400792 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500793 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400794 nd->inode = nd->path.dentry->d_inode;
795 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400796 /* stepped on a _really_ weird one */
797 path_put(&nd->path);
798 error = -ELOOP;
799 }
800 }
Ian Kent051d3812006-03-27 01:14:53 -0800801 }
Ian Kent051d3812006-03-27 01:14:53 -0800802 return error;
803}
804
Nick Piggin31e6b012011-01-07 17:49:52 +1100805static int follow_up_rcu(struct path *path)
806{
807 struct vfsmount *parent;
808 struct dentry *mountpoint;
809
810 parent = path->mnt->mnt_parent;
811 if (parent == path->mnt)
812 return 0;
813 mountpoint = path->mnt->mnt_mountpoint;
814 path->dentry = mountpoint;
815 path->mnt = parent;
816 return 1;
817}
818
Al Virobab77eb2009-04-18 03:26:48 -0400819int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
821 struct vfsmount *parent;
822 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000823
824 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400825 parent = path->mnt->mnt_parent;
826 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000827 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return 0;
829 }
830 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400831 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000832 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400833 dput(path->dentry);
834 path->dentry = mountpoint;
835 mntput(path->mnt);
836 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return 1;
838}
839
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100840/*
David Howells9875cf82011-01-14 18:45:21 +0000841 * Perform an automount
842 * - return -EISDIR to tell follow_managed() to stop and return the path we
843 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 */
David Howells9875cf82011-01-14 18:45:21 +0000845static int follow_automount(struct path *path, unsigned flags,
846 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100847{
David Howells9875cf82011-01-14 18:45:21 +0000848 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000849 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100850
David Howells9875cf82011-01-14 18:45:21 +0000851 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
852 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700853
David Howells6f45b652011-01-14 18:45:31 +0000854 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
855 * and this is the terminal part of the path.
856 */
857 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
858 return -EISDIR; /* we actually want to stop here */
859
David Howells9875cf82011-01-14 18:45:21 +0000860 /* We want to mount if someone is trying to open/create a file of any
861 * type under the mountpoint, wants to traverse through the mountpoint
862 * or wants to open the mounted directory.
863 *
864 * We don't want to mount if someone's just doing a stat and they've
865 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
866 * appended a '/' to the name.
867 */
868 if (!(flags & LOOKUP_FOLLOW) &&
869 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
870 LOOKUP_OPEN | LOOKUP_CREATE)))
871 return -EISDIR;
872
873 current->total_link_count++;
874 if (current->total_link_count >= 40)
875 return -ELOOP;
876
877 mnt = path->dentry->d_op->d_automount(path);
878 if (IS_ERR(mnt)) {
879 /*
880 * The filesystem is allowed to return -EISDIR here to indicate
881 * it doesn't want to automount. For instance, autofs would do
882 * this so that its userspace daemon can mount on this dentry.
883 *
884 * However, we can only permit this if it's a terminal point in
885 * the path being looked up; if it wasn't then the remainder of
886 * the path is inaccessible and we should say so.
887 */
888 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
889 return -EREMOTE;
890 return PTR_ERR(mnt);
891 }
David Howellsea5b7782011-01-14 19:10:03 +0000892
David Howells9875cf82011-01-14 18:45:21 +0000893 if (!mnt) /* mount collision */
894 return 0;
895
Al Viro19a167a2011-01-17 01:35:23 -0500896 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000897
David Howellsea5b7782011-01-14 19:10:03 +0000898 switch (err) {
899 case -EBUSY:
900 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500901 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000902 case 0:
David Howellsea5b7782011-01-14 19:10:03 +0000903 dput(path->dentry);
904 if (*need_mntput)
905 mntput(path->mnt);
906 path->mnt = mnt;
907 path->dentry = dget(mnt->mnt_root);
908 *need_mntput = true;
909 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500910 default:
911 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000912 }
Al Viro19a167a2011-01-17 01:35:23 -0500913
David Howells9875cf82011-01-14 18:45:21 +0000914}
915
916/*
917 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000918 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000919 * - Flagged as mountpoint
920 * - Flagged as automount point
921 *
922 * This may only be called in refwalk mode.
923 *
924 * Serialization is taken care of in namespace.c
925 */
926static int follow_managed(struct path *path, unsigned flags)
927{
928 unsigned managed;
929 bool need_mntput = false;
930 int ret;
931
932 /* Given that we're not holding a lock here, we retain the value in a
933 * local variable for each dentry as we look at it so that we don't see
934 * the components of that value change under us */
935 while (managed = ACCESS_ONCE(path->dentry->d_flags),
936 managed &= DCACHE_MANAGED_DENTRY,
937 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000938 /* Allow the filesystem to manage the transit without i_mutex
939 * being held. */
940 if (managed & DCACHE_MANAGE_TRANSIT) {
941 BUG_ON(!path->dentry->d_op);
942 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400943 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000944 if (ret < 0)
945 return ret == -EISDIR ? 0 : ret;
946 }
947
David Howells9875cf82011-01-14 18:45:21 +0000948 /* Transit to a mounted filesystem. */
949 if (managed & DCACHE_MOUNTED) {
950 struct vfsmount *mounted = lookup_mnt(path);
951 if (mounted) {
952 dput(path->dentry);
953 if (need_mntput)
954 mntput(path->mnt);
955 path->mnt = mounted;
956 path->dentry = dget(mounted->mnt_root);
957 need_mntput = true;
958 continue;
959 }
960
961 /* Something is mounted on this dentry in another
962 * namespace and/or whatever was mounted there in this
963 * namespace got unmounted before we managed to get the
964 * vfsmount_lock */
965 }
966
967 /* Handle an automount point */
968 if (managed & DCACHE_NEED_AUTOMOUNT) {
969 ret = follow_automount(path, flags, &need_mntput);
970 if (ret < 0)
971 return ret == -EISDIR ? 0 : ret;
972 continue;
973 }
974
975 /* We didn't change the current path point */
976 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 }
David Howells9875cf82011-01-14 18:45:21 +0000978 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979}
980
David Howellscc53ce52011-01-14 18:45:26 +0000981int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
983 struct vfsmount *mounted;
984
Al Viro1c755af2009-04-18 14:06:57 -0400985 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400987 dput(path->dentry);
988 mntput(path->mnt);
989 path->mnt = mounted;
990 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 return 1;
992 }
993 return 0;
994}
995
Ian Kent62a73752011-03-25 01:51:02 +0800996static inline bool managed_dentry_might_block(struct dentry *dentry)
997{
998 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
999 dentry->d_op->d_manage(dentry, true) < 0);
1000}
1001
David Howells9875cf82011-01-14 18:45:21 +00001002/*
1003 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
David Howellscc53ce52011-01-14 18:45:26 +00001004 * meet a managed dentry and we're not walking to "..". True is returned to
David Howells9875cf82011-01-14 18:45:21 +00001005 * continue, false to abort.
1006 */
1007static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1008 struct inode **inode, bool reverse_transit)
1009{
Ian Kent62a73752011-03-25 01:51:02 +08001010 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +00001011 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001012 /*
1013 * Don't forget we might have a non-mountpoint managed dentry
1014 * that wants to block transit.
1015 */
1016 *inode = path->dentry->d_inode;
1017 if (!reverse_transit &&
1018 unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001019 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001020
1021 if (!d_mountpoint(path->dentry))
1022 break;
1023
David Howells9875cf82011-01-14 18:45:21 +00001024 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1025 if (!mounted)
1026 break;
1027 path->mnt = mounted;
1028 path->dentry = mounted->mnt_root;
1029 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
David Howells9875cf82011-01-14 18:45:21 +00001030 }
1031
1032 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1033 return reverse_transit;
1034 return true;
1035}
1036
Nick Piggin31e6b012011-01-07 17:49:52 +11001037static int follow_dotdot_rcu(struct nameidata *nd)
1038{
1039 struct inode *inode = nd->inode;
1040
1041 set_root_rcu(nd);
1042
David Howells9875cf82011-01-14 18:45:21 +00001043 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001044 if (nd->path.dentry == nd->root.dentry &&
1045 nd->path.mnt == nd->root.mnt) {
1046 break;
1047 }
1048 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1049 struct dentry *old = nd->path.dentry;
1050 struct dentry *parent = old->d_parent;
1051 unsigned seq;
1052
1053 seq = read_seqcount_begin(&parent->d_seq);
1054 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001055 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001056 inode = parent->d_inode;
1057 nd->path.dentry = parent;
1058 nd->seq = seq;
1059 break;
1060 }
1061 if (!follow_up_rcu(&nd->path))
1062 break;
1063 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1064 inode = nd->path.dentry->d_inode;
1065 }
David Howells9875cf82011-01-14 18:45:21 +00001066 __follow_mount_rcu(nd, &nd->path, &inode, true);
Nick Piggin31e6b012011-01-07 17:49:52 +11001067 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001068 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001069
1070failed:
1071 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001072 if (!(nd->flags & LOOKUP_ROOT))
1073 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -05001074 rcu_read_unlock();
1075 br_read_unlock(vfsmount_lock);
1076 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001077}
1078
David Howells9875cf82011-01-14 18:45:21 +00001079/*
David Howellscc53ce52011-01-14 18:45:26 +00001080 * Follow down to the covering mount currently visible to userspace. At each
1081 * point, the filesystem owning that dentry may be queried as to whether the
1082 * caller is permitted to proceed or not.
1083 *
1084 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1085 * being true).
1086 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001087int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001088{
1089 unsigned managed;
1090 int ret;
1091
1092 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1093 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1094 /* Allow the filesystem to manage the transit without i_mutex
1095 * being held.
1096 *
1097 * We indicate to the filesystem if someone is trying to mount
1098 * something here. This gives autofs the chance to deny anyone
1099 * other than its daemon the right to mount on its
1100 * superstructure.
1101 *
1102 * The filesystem may sleep at this point.
1103 */
1104 if (managed & DCACHE_MANAGE_TRANSIT) {
1105 BUG_ON(!path->dentry->d_op);
1106 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001107 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001108 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001109 if (ret < 0)
1110 return ret == -EISDIR ? 0 : ret;
1111 }
1112
1113 /* Transit to a mounted filesystem. */
1114 if (managed & DCACHE_MOUNTED) {
1115 struct vfsmount *mounted = lookup_mnt(path);
1116 if (!mounted)
1117 break;
1118 dput(path->dentry);
1119 mntput(path->mnt);
1120 path->mnt = mounted;
1121 path->dentry = dget(mounted->mnt_root);
1122 continue;
1123 }
1124
1125 /* Don't handle automount points here */
1126 break;
1127 }
1128 return 0;
1129}
1130
1131/*
David Howells9875cf82011-01-14 18:45:21 +00001132 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1133 */
1134static void follow_mount(struct path *path)
1135{
1136 while (d_mountpoint(path->dentry)) {
1137 struct vfsmount *mounted = lookup_mnt(path);
1138 if (!mounted)
1139 break;
1140 dput(path->dentry);
1141 mntput(path->mnt);
1142 path->mnt = mounted;
1143 path->dentry = dget(mounted->mnt_root);
1144 }
1145}
1146
Nick Piggin31e6b012011-01-07 17:49:52 +11001147static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
Al Viro2a737872009-04-07 11:49:53 -04001149 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001152 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
Al Viro2a737872009-04-07 11:49:53 -04001154 if (nd->path.dentry == nd->root.dentry &&
1155 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 break;
1157 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001158 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001159 /* rare case of legitimate dget_parent()... */
1160 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 dput(old);
1162 break;
1163 }
Al Viro3088dd72010-01-30 15:47:29 -05001164 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Al Viro79ed0222009-04-18 13:59:41 -04001167 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001168 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169}
1170
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001172 * Allocate a dentry with name and parent, and perform a parent
1173 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1174 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1175 * have verified that no child exists while under i_mutex.
1176 */
1177static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1178 struct qstr *name, struct nameidata *nd)
1179{
1180 struct inode *inode = parent->d_inode;
1181 struct dentry *dentry;
1182 struct dentry *old;
1183
1184 /* Don't create child dentry for a dead directory. */
1185 if (unlikely(IS_DEADDIR(inode)))
1186 return ERR_PTR(-ENOENT);
1187
1188 dentry = d_alloc(parent, name);
1189 if (unlikely(!dentry))
1190 return ERR_PTR(-ENOMEM);
1191
1192 old = inode->i_op->lookup(inode, dentry, nd);
1193 if (unlikely(old)) {
1194 dput(dentry);
1195 dentry = old;
1196 }
1197 return dentry;
1198}
1199
1200/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 * It's more convoluted than I'd like it to be, but... it's still fairly
1202 * small and for now I'd prefer to have fast path as straight as possible.
1203 * It _is_ time-critical.
1204 */
1205static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001206 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207{
Jan Blunck4ac91372008-02-14 19:34:32 -08001208 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001209 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001210 int need_reval = 1;
1211 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001212 int err;
1213
Al Viro3cac2602009-08-13 18:27:43 +04001214 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001215 * Rename seqlock is not required here because in the off chance
1216 * of a false negative due to a concurrent rename, we're going to
1217 * do the non-racy lookup, below.
1218 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001219 if (nd->flags & LOOKUP_RCU) {
1220 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001221 *inode = nd->inode;
1222 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001223 if (!dentry)
1224 goto unlazy;
1225
Nick Piggin31e6b012011-01-07 17:49:52 +11001226 /* Memory barrier in read_seqcount_begin of child is enough */
1227 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1228 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001229 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001230
Al Viro24643082011-02-15 01:26:22 -05001231 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001232 status = d_revalidate(dentry, nd);
1233 if (unlikely(status <= 0)) {
1234 if (status != -ECHILD)
1235 need_reval = 0;
1236 goto unlazy;
1237 }
Al Viro24643082011-02-15 01:26:22 -05001238 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001239 path->mnt = mnt;
1240 path->dentry = dentry;
David Howells9875cf82011-01-14 18:45:21 +00001241 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1242 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001243unlazy:
1244 if (dentry) {
1245 if (nameidata_dentry_drop_rcu(nd, dentry))
1246 return -ECHILD;
1247 } else {
1248 if (nameidata_drop_rcu(nd))
1249 return -ECHILD;
1250 }
1251 } else {
1252 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001253 }
Al Viro5a18fff2011-03-11 04:44:53 -05001254
1255retry:
1256 if (unlikely(!dentry)) {
1257 struct inode *dir = parent->d_inode;
1258 BUG_ON(nd->inode != dir);
1259
1260 mutex_lock(&dir->i_mutex);
1261 dentry = d_lookup(parent, name);
1262 if (likely(!dentry)) {
1263 dentry = d_alloc_and_lookup(parent, name, nd);
1264 if (IS_ERR(dentry)) {
1265 mutex_unlock(&dir->i_mutex);
1266 return PTR_ERR(dentry);
1267 }
1268 /* known good */
1269 need_reval = 0;
1270 status = 1;
1271 }
1272 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001273 }
Al Viro5a18fff2011-03-11 04:44:53 -05001274 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1275 status = d_revalidate(dentry, nd);
1276 if (unlikely(status <= 0)) {
1277 if (status < 0) {
1278 dput(dentry);
1279 return status;
1280 }
1281 if (!d_invalidate(dentry)) {
1282 dput(dentry);
1283 dentry = NULL;
1284 need_reval = 1;
1285 goto retry;
1286 }
1287 }
1288
David Howells9875cf82011-01-14 18:45:21 +00001289 path->mnt = mnt;
1290 path->dentry = dentry;
1291 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001292 if (unlikely(err < 0)) {
1293 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001294 return err;
Ian Kent89312212011-01-18 12:06:10 +08001295 }
David Howells9875cf82011-01-14 18:45:21 +00001296 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
1299
Al Viro52094c82011-02-21 21:34:47 -05001300static inline int may_lookup(struct nameidata *nd)
1301{
1302 if (nd->flags & LOOKUP_RCU) {
1303 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1304 if (err != -ECHILD)
1305 return err;
1306 if (nameidata_drop_rcu(nd))
1307 return -ECHILD;
1308 }
1309 return exec_permission(nd->inode, 0);
1310}
1311
Al Viro9856fa12011-03-04 14:22:06 -05001312static inline int handle_dots(struct nameidata *nd, int type)
1313{
1314 if (type == LAST_DOTDOT) {
1315 if (nd->flags & LOOKUP_RCU) {
1316 if (follow_dotdot_rcu(nd))
1317 return -ECHILD;
1318 } else
1319 follow_dotdot(nd);
1320 }
1321 return 0;
1322}
1323
Al Viro951361f2011-03-04 14:44:37 -05001324static void terminate_walk(struct nameidata *nd)
1325{
1326 if (!(nd->flags & LOOKUP_RCU)) {
1327 path_put(&nd->path);
1328 } else {
1329 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001330 if (!(nd->flags & LOOKUP_ROOT))
1331 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001332 rcu_read_unlock();
1333 br_read_unlock(vfsmount_lock);
1334 }
1335}
1336
Al Viroce57dfc2011-03-13 19:58:58 -04001337static inline int walk_component(struct nameidata *nd, struct path *path,
1338 struct qstr *name, int type, int follow)
1339{
1340 struct inode *inode;
1341 int err;
1342 /*
1343 * "." and ".." are special - ".." especially so because it has
1344 * to be able to know about the current root directory and
1345 * parent relationships.
1346 */
1347 if (unlikely(type != LAST_NORM))
1348 return handle_dots(nd, type);
1349 err = do_lookup(nd, name, path, &inode);
1350 if (unlikely(err)) {
1351 terminate_walk(nd);
1352 return err;
1353 }
1354 if (!inode) {
1355 path_to_nameidata(path, nd);
1356 terminate_walk(nd);
1357 return -ENOENT;
1358 }
1359 if (unlikely(inode->i_op->follow_link) && follow) {
1360 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1361 return -ECHILD;
1362 BUG_ON(inode != path->dentry->d_inode);
1363 return 1;
1364 }
1365 path_to_nameidata(path, nd);
1366 nd->inode = inode;
1367 return 0;
1368}
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370/*
Al Virob3563792011-03-14 21:54:55 -04001371 * This limits recursive symlink follows to 8, while
1372 * limiting consecutive symlinks to 40.
1373 *
1374 * Without that kind of total limit, nasty chains of consecutive
1375 * symlinks can cause almost arbitrarily long lookups.
1376 */
1377static inline int nested_symlink(struct path *path, struct nameidata *nd)
1378{
1379 int res;
1380
Al Virob3563792011-03-14 21:54:55 -04001381 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1382 path_put_conditional(path, nd);
1383 path_put(&nd->path);
1384 return -ELOOP;
1385 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001386 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001387
1388 nd->depth++;
1389 current->link_count++;
1390
1391 do {
1392 struct path link = *path;
1393 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001394
1395 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001396 if (!res)
1397 res = walk_component(nd, path, &nd->last,
1398 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001399 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001400 } while (res > 0);
1401
1402 current->link_count--;
1403 nd->depth--;
1404 return res;
1405}
1406
1407/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001409 * This is the basic name resolution function, turning a pathname into
1410 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001412 * Returns 0 and nd will have valid dentry and mnt on success.
1413 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 */
Al Viro6de88d72009-08-09 01:41:57 +04001415static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416{
1417 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 int err;
1419 unsigned int lookup_flags = nd->flags;
1420
1421 while (*name=='/')
1422 name++;
1423 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001424 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 /* At this point we know we have a real path component. */
1427 for(;;) {
1428 unsigned long hash;
1429 struct qstr this;
1430 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001431 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Trond Myklebustcdce5d62005-10-18 14:20:18 -07001433 nd->flags |= LOOKUP_CONTINUE;
Al Viro52094c82011-02-21 21:34:47 -05001434
1435 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (err)
1437 break;
1438
1439 this.name = name;
1440 c = *(const unsigned char *)name;
1441
1442 hash = init_name_hash();
1443 do {
1444 name++;
1445 hash = partial_name_hash(c, hash);
1446 c = *(const unsigned char *)name;
1447 } while (c && (c != '/'));
1448 this.len = name - (const char *) this.name;
1449 this.hash = end_name_hash(hash);
1450
Al Virofe479a52011-02-22 15:10:03 -05001451 type = LAST_NORM;
1452 if (this.name[0] == '.') switch (this.len) {
1453 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001454 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001455 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001456 nd->flags |= LOOKUP_JUMPED;
1457 }
Al Virofe479a52011-02-22 15:10:03 -05001458 break;
1459 case 1:
1460 type = LAST_DOT;
1461 }
Al Viro5a202bc2011-03-08 14:17:44 -05001462 if (likely(type == LAST_NORM)) {
1463 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001464 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001465 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1466 err = parent->d_op->d_hash(parent, nd->inode,
1467 &this);
1468 if (err < 0)
1469 break;
1470 }
1471 }
Al Virofe479a52011-02-22 15:10:03 -05001472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 /* remove trailing slashes? */
1474 if (!c)
1475 goto last_component;
1476 while (*++name == '/');
1477 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001478 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Al Viroce57dfc2011-03-13 19:58:58 -04001480 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1481 if (err < 0)
1482 return err;
Al Virofe479a52011-02-22 15:10:03 -05001483
Al Viroce57dfc2011-03-13 19:58:58 -04001484 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001485 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001487 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001488 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001490 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 break;
1492 continue;
1493 /* here ends the main loop */
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -08001496 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1497 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Al Virob3563792011-03-14 21:54:55 -04001498 nd->last = this;
1499 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001500 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 }
Al Viro951361f2011-03-04 14:44:37 -05001502 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return err;
1504}
1505
Al Viro70e9b352011-03-05 21:12:22 -05001506static int path_init(int dfd, const char *name, unsigned int flags,
1507 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001509 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001510 int fput_needed;
1511 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
1513 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001514 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001516 if (flags & LOOKUP_ROOT) {
1517 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001518 if (*name) {
1519 if (!inode->i_op->lookup)
1520 return -ENOTDIR;
1521 retval = inode_permission(inode, MAY_EXEC);
1522 if (retval)
1523 return retval;
1524 }
Al Viro5b6ca022011-03-09 23:04:47 -05001525 nd->path = nd->root;
1526 nd->inode = inode;
1527 if (flags & LOOKUP_RCU) {
1528 br_read_lock(vfsmount_lock);
1529 rcu_read_lock();
1530 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1531 } else {
1532 path_get(&nd->path);
1533 }
1534 return 0;
1535 }
1536
Al Viro2a737872009-04-07 11:49:53 -04001537 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001540 if (flags & LOOKUP_RCU) {
1541 br_read_lock(vfsmount_lock);
1542 rcu_read_lock();
1543 set_root_rcu(nd);
1544 } else {
1545 set_root(nd);
1546 path_get(&nd->root);
1547 }
Al Viro2a737872009-04-07 11:49:53 -04001548 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001549 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001550 if (flags & LOOKUP_RCU) {
1551 struct fs_struct *fs = current->fs;
1552 unsigned seq;
1553
1554 br_read_lock(vfsmount_lock);
1555 rcu_read_lock();
1556
1557 do {
1558 seq = read_seqcount_begin(&fs->seq);
1559 nd->path = fs->pwd;
1560 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1561 } while (read_seqcount_retry(&fs->seq, seq));
1562 } else {
1563 get_fs_pwd(current->fs, &nd->path);
1564 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001565 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001566 struct dentry *dentry;
1567
Al Viro1abf0c72011-03-13 03:51:11 -04001568 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001569 retval = -EBADF;
1570 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001571 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001572
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001573 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001574
Al Virof52e0c12011-03-14 18:56:51 -04001575 if (*name) {
1576 retval = -ENOTDIR;
1577 if (!S_ISDIR(dentry->d_inode->i_mode))
1578 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001579
Al Virof52e0c12011-03-14 18:56:51 -04001580 retval = file_permission(file, MAY_EXEC);
1581 if (retval)
1582 goto fput_fail;
1583 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001584
Jan Blunck5dd784d02008-02-14 19:34:38 -08001585 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001586 if (flags & LOOKUP_RCU) {
1587 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001588 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001589 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1590 br_read_lock(vfsmount_lock);
1591 rcu_read_lock();
1592 } else {
1593 path_get(&file->f_path);
1594 fput_light(file, fput_needed);
1595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
Al Viroe41f7d42011-02-22 14:02:58 -05001597
Nick Piggin31e6b012011-01-07 17:49:52 +11001598 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001599 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001600
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001601fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001602 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001603out_fail:
1604 return retval;
1605}
1606
Al Virobd92d7f2011-03-14 19:54:59 -04001607static inline int lookup_last(struct nameidata *nd, struct path *path)
1608{
1609 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1610 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1611
1612 nd->flags &= ~LOOKUP_PARENT;
1613 return walk_component(nd, path, &nd->last, nd->last_type,
1614 nd->flags & LOOKUP_FOLLOW);
1615}
1616
Al Viro9b4a9b12009-04-07 11:44:16 -04001617/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001618static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001619 unsigned int flags, struct nameidata *nd)
1620{
Al Viro70e9b352011-03-05 21:12:22 -05001621 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001622 struct path path;
1623 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001624
1625 /*
1626 * Path walking is largely split up into 2 different synchronisation
1627 * schemes, rcu-walk and ref-walk (explained in
1628 * Documentation/filesystems/path-lookup.txt). These share much of the
1629 * path walk code, but some things particularly setup, cleanup, and
1630 * following mounts are sufficiently divergent that functions are
1631 * duplicated. Typically there is a function foo(), and its RCU
1632 * analogue, foo_rcu().
1633 *
1634 * -ECHILD is the error number of choice (just to avoid clashes) that
1635 * is returned if some aspect of an rcu-walk fails. Such an error must
1636 * be handled by restarting a traditional ref-walk (which will always
1637 * be able to complete).
1638 */
Al Virobd92d7f2011-03-14 19:54:59 -04001639 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001640
Al Virobd92d7f2011-03-14 19:54:59 -04001641 if (unlikely(err))
1642 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001643
1644 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001645 err = link_path_walk(name, nd);
1646
1647 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001648 err = lookup_last(nd, &path);
1649 while (err > 0) {
1650 void *cookie;
1651 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001652 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001653 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001654 if (!err)
1655 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001656 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001657 }
1658 }
Al Viroee0827c2011-02-21 23:38:09 -05001659
1660 if (nd->flags & LOOKUP_RCU) {
Al Viro4455ca62011-03-04 14:28:10 -05001661 /* went all way through without dropping RCU */
Al Virobd92d7f2011-03-14 19:54:59 -04001662 BUG_ON(err);
Al Viro4455ca62011-03-04 14:28:10 -05001663 if (nameidata_drop_rcu_last(nd))
Al Virobd92d7f2011-03-14 19:54:59 -04001664 err = -ECHILD;
Al Viroee0827c2011-02-21 23:38:09 -05001665 }
1666
Al Virobd23a532011-03-23 09:56:30 -04001667 if (!err) {
Al Virobd92d7f2011-03-14 19:54:59 -04001668 err = handle_reval_path(nd);
Al Virobd23a532011-03-23 09:56:30 -04001669 if (err)
1670 path_put(&nd->path);
1671 }
Al Virobd92d7f2011-03-14 19:54:59 -04001672
1673 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1674 if (!nd->inode->i_op->lookup) {
1675 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001676 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001677 }
1678 }
Al Viro16c2cd72011-02-22 15:50:10 -05001679
Al Viro70e9b352011-03-05 21:12:22 -05001680 if (base)
1681 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001682
Al Viro5b6ca022011-03-09 23:04:47 -05001683 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001684 path_put(&nd->root);
1685 nd->root.mnt = NULL;
1686 }
Al Virobd92d7f2011-03-14 19:54:59 -04001687 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001688}
Nick Piggin31e6b012011-01-07 17:49:52 +11001689
Al Viroee0827c2011-02-21 23:38:09 -05001690static int do_path_lookup(int dfd, const char *name,
1691 unsigned int flags, struct nameidata *nd)
1692{
1693 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1694 if (unlikely(retval == -ECHILD))
1695 retval = path_lookupat(dfd, name, flags, nd);
1696 if (unlikely(retval == -ESTALE))
1697 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001698
1699 if (likely(!retval)) {
1700 if (unlikely(!audit_dummy_context())) {
1701 if (nd->path.dentry && nd->inode)
1702 audit_inode(name, nd->path.dentry);
1703 }
1704 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001705 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706}
1707
Al Viroc9c6cac2011-02-16 15:15:47 -05001708int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001709{
Al Viroc9c6cac2011-02-16 15:15:47 -05001710 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001711}
1712
Al Virod1811462008-08-02 00:49:18 -04001713int kern_path(const char *name, unsigned int flags, struct path *path)
1714{
1715 struct nameidata nd;
1716 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1717 if (!res)
1718 *path = nd.path;
1719 return res;
1720}
1721
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001722/**
1723 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1724 * @dentry: pointer to dentry of the base directory
1725 * @mnt: pointer to vfs mount of the base directory
1726 * @name: pointer to file name
1727 * @flags: lookup flags
1728 * @nd: pointer to nameidata
1729 */
1730int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1731 const char *name, unsigned int flags,
1732 struct nameidata *nd)
1733{
Al Viro5b6ca022011-03-09 23:04:47 -05001734 nd->root.dentry = dentry;
1735 nd->root.mnt = mnt;
1736 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1737 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001738}
1739
Christoph Hellwigeead1912007-10-16 23:25:38 -07001740static struct dentry *__lookup_hash(struct qstr *name,
1741 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001743 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001744 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 int err;
1746
Nick Pigginb74c79e2011-01-07 17:49:58 +11001747 err = exec_permission(inode, 0);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001748 if (err)
1749 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
1751 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001752 * Don't bother with __d_lookup: callers are for creat as
1753 * well as unlink, so a lot of the time it would cost
1754 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001755 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001756 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001757
Nick Pigginfb045ad2011-01-07 17:49:55 +11001758 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro6e6b1bd2009-08-13 23:38:37 +04001759 dentry = do_revalidate(dentry, nd);
1760
Nick Pigginbaa03892010-08-18 04:37:31 +10001761 if (!dentry)
1762 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001763
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 return dentry;
1765}
1766
James Morris057f6c02007-04-26 00:12:05 -07001767/*
1768 * Restricted form of lookup. Doesn't follow links, single-component only,
1769 * needs parent already locked. Doesn't follow mounts.
1770 * SMP-safe.
1771 */
Adrian Bunka244e162006-03-31 02:32:11 -08001772static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773{
Jan Blunck4ac91372008-02-14 19:34:32 -08001774 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775}
1776
Christoph Hellwigeead1912007-10-16 23:25:38 -07001777/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001778 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001779 * @name: pathname component to lookup
1780 * @base: base directory to lookup from
1781 * @len: maximum length @len should be interpreted to
1782 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001783 * Note that this routine is purely a helper for filesystem usage and should
1784 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001785 * nameidata argument is passed to the filesystem methods and a filesystem
1786 * using this helper needs to be prepared for that.
1787 */
James Morris057f6c02007-04-26 00:12:05 -07001788struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1789{
James Morris057f6c02007-04-26 00:12:05 -07001790 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001791 unsigned long hash;
1792 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001793
David Woodhouse2f9092e2009-04-20 23:18:37 +01001794 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1795
Al Viro6a96ba52011-03-07 23:49:20 -05001796 this.name = name;
1797 this.len = len;
1798 if (!len)
1799 return ERR_PTR(-EACCES);
1800
1801 hash = init_name_hash();
1802 while (len--) {
1803 c = *(const unsigned char *)name++;
1804 if (c == '/' || c == '\0')
1805 return ERR_PTR(-EACCES);
1806 hash = partial_name_hash(c, hash);
1807 }
1808 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001809 /*
1810 * See if the low-level filesystem might want
1811 * to use its own hash..
1812 */
1813 if (base->d_flags & DCACHE_OP_HASH) {
1814 int err = base->d_op->d_hash(base, base->d_inode, &this);
1815 if (err < 0)
1816 return ERR_PTR(err);
1817 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001818
Christoph Hellwig49705b72005-11-08 21:35:06 -08001819 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001820}
1821
Al Viro2d8f3032008-07-22 09:59:21 -04001822int user_path_at(int dfd, const char __user *name, unsigned flags,
1823 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824{
Al Viro2d8f3032008-07-22 09:59:21 -04001825 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001826 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001829
1830 BUG_ON(flags & LOOKUP_PARENT);
1831
1832 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001834 if (!err)
1835 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 }
1837 return err;
1838}
1839
Al Viro2ad94ae2008-07-21 09:32:51 -04001840static int user_path_parent(int dfd, const char __user *path,
1841 struct nameidata *nd, char **name)
1842{
1843 char *s = getname(path);
1844 int error;
1845
1846 if (IS_ERR(s))
1847 return PTR_ERR(s);
1848
1849 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1850 if (error)
1851 putname(s);
1852 else
1853 *name = s;
1854
1855 return error;
1856}
1857
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858/*
1859 * It's inline, so penalty for filesystems that don't use sticky bit is
1860 * minimal.
1861 */
1862static inline int check_sticky(struct inode *dir, struct inode *inode)
1863{
David Howellsda9592e2008-11-14 10:39:05 +11001864 uid_t fsuid = current_fsuid();
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 if (!(dir->i_mode & S_ISVTX))
1867 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001868 if (current_user_ns() != inode_userns(inode))
1869 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001870 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001872 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001874
1875other_userns:
1876 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877}
1878
1879/*
1880 * Check whether we can remove a link victim from directory dir, check
1881 * whether the type of victim is right.
1882 * 1. We can't do it if dir is read-only (done in permission())
1883 * 2. We should have write and exec permissions on dir
1884 * 3. We can't remove anything from append-only dir
1885 * 4. We can't do anything with immutable dir (done in permission())
1886 * 5. If the sticky bit on dir is set we should either
1887 * a. be owner of dir, or
1888 * b. be owner of victim, or
1889 * c. have CAP_FOWNER capability
1890 * 6. If the victim is append-only or immutable we can't do antyhing with
1891 * links pointing to it.
1892 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1893 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1894 * 9. We can't remove a root or mountpoint.
1895 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1896 * nfs_async_unlink().
1897 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001898static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899{
1900 int error;
1901
1902 if (!victim->d_inode)
1903 return -ENOENT;
1904
1905 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001906 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Al Virof419a2e2008-07-22 00:07:17 -04001908 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (error)
1910 return error;
1911 if (IS_APPEND(dir))
1912 return -EPERM;
1913 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001914 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 return -EPERM;
1916 if (isdir) {
1917 if (!S_ISDIR(victim->d_inode->i_mode))
1918 return -ENOTDIR;
1919 if (IS_ROOT(victim))
1920 return -EBUSY;
1921 } else if (S_ISDIR(victim->d_inode->i_mode))
1922 return -EISDIR;
1923 if (IS_DEADDIR(dir))
1924 return -ENOENT;
1925 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1926 return -EBUSY;
1927 return 0;
1928}
1929
1930/* Check whether we can create an object with dentry child in directory
1931 * dir.
1932 * 1. We can't do it if child already exists (open has special treatment for
1933 * this case, but since we are inlined it's OK)
1934 * 2. We can't do it if dir is read-only (done in permission())
1935 * 3. We should have write and exec permissions on dir
1936 * 4. We can't do it if dir is immutable (done in permission())
1937 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001938static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939{
1940 if (child->d_inode)
1941 return -EEXIST;
1942 if (IS_DEADDIR(dir))
1943 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001944 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945}
1946
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947/*
1948 * p1 and p2 should be directories on the same fs.
1949 */
1950struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1951{
1952 struct dentry *p;
1953
1954 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001955 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 return NULL;
1957 }
1958
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001959 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001961 p = d_ancestor(p2, p1);
1962 if (p) {
1963 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1964 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1965 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 }
1967
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001968 p = d_ancestor(p1, p2);
1969 if (p) {
1970 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1971 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1972 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 }
1974
Ingo Molnarf2eace22006-07-03 00:25:05 -07001975 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1976 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 return NULL;
1978}
1979
1980void unlock_rename(struct dentry *p1, struct dentry *p2)
1981{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001982 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001984 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001985 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 }
1987}
1988
1989int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1990 struct nameidata *nd)
1991{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001992 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
1994 if (error)
1995 return error;
1996
Al Viroacfa4382008-12-04 10:06:33 -05001997 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 return -EACCES; /* shouldn't it be ENOSYS? */
1999 mode &= S_IALLUGO;
2000 mode |= S_IFREG;
2001 error = security_inode_create(dir, dentry, mode);
2002 if (error)
2003 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002005 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002006 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 return error;
2008}
2009
Al Viro73d049a2011-03-11 12:08:24 -05002010static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002012 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 struct inode *inode = dentry->d_inode;
2014 int error;
2015
Al Virobcda7652011-03-13 16:42:14 -04002016 /* O_PATH? */
2017 if (!acc_mode)
2018 return 0;
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 if (!inode)
2021 return -ENOENT;
2022
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002023 switch (inode->i_mode & S_IFMT) {
2024 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002026 case S_IFDIR:
2027 if (acc_mode & MAY_WRITE)
2028 return -EISDIR;
2029 break;
2030 case S_IFBLK:
2031 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002032 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002034 /*FALLTHRU*/
2035 case S_IFIFO:
2036 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002038 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002039 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002040
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002041 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002042 if (error)
2043 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 /*
2046 * An append-only file must be opened in append mode for writing.
2047 */
2048 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002049 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002050 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002052 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 }
2054
2055 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002056 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002057 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
2059 /*
2060 * Ensure there are no outstanding leases on the file.
2061 */
Al Virob65a9cf2009-12-16 06:27:40 -05002062 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002063}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Jeff Laytone1181ee2010-12-07 16:19:50 -05002065static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002066{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002067 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002068 struct inode *inode = path->dentry->d_inode;
2069 int error = get_write_access(inode);
2070 if (error)
2071 return error;
2072 /*
2073 * Refuse to truncate files with mandatory locks held on them.
2074 */
2075 error = locks_verify_locked(inode);
2076 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002077 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002078 if (!error) {
2079 error = do_truncate(path->dentry, 0,
2080 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002081 filp);
Al Viro7715b522009-12-16 03:54:00 -05002082 }
2083 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002084 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085}
2086
Dave Hansend57999e2008-02-15 14:37:27 -08002087/*
Dave Hansend57999e2008-02-15 14:37:27 -08002088 * Note that while the flag value (low two bits) for sys_open means:
2089 * 00 - read-only
2090 * 01 - write-only
2091 * 10 - read-write
2092 * 11 - special
2093 * it is changed into
2094 * 00 - no permissions needed
2095 * 01 - read-permission
2096 * 10 - write-permission
2097 * 11 - read-write
2098 * for the internal routines (ie open_namei()/follow_link() etc)
2099 * This is more logical, and also allows the 00 "no perm needed"
2100 * to be used for symlinks (where the permissions are checked
2101 * later).
2102 *
2103*/
2104static inline int open_to_namei_flags(int flag)
2105{
2106 if ((flag+1) & O_ACCMODE)
2107 flag++;
2108 return flag;
2109}
2110
Nick Piggin31e6b012011-01-07 17:49:52 +11002111/*
Al Virofe2d35f2011-03-05 22:58:25 -05002112 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002113 */
Al Virofb1cc552009-12-24 01:58:28 -05002114static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002115 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002116{
Al Viroa1e28032009-12-24 02:12:06 -05002117 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002118 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002119 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002120 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002121 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002122 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002123 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002124 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002125
Al Viroc3e380b2011-02-23 13:39:45 -05002126 nd->flags &= ~LOOKUP_PARENT;
2127 nd->flags |= op->intent;
2128
Al Viro1f36f772009-12-26 10:56:19 -05002129 switch (nd->last_type) {
2130 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002131 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002132 error = handle_dots(nd, nd->last_type);
2133 if (error)
2134 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002135 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002136 case LAST_ROOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002137 if (nd->flags & LOOKUP_RCU) {
2138 if (nameidata_drop_rcu_last(nd))
2139 return ERR_PTR(-ECHILD);
2140 }
Al Viro16c2cd72011-02-22 15:50:10 -05002141 error = handle_reval_path(nd);
2142 if (error)
2143 goto exit;
Al Virofe2d35f2011-03-05 22:58:25 -05002144 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002145 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002146 error = -EISDIR;
2147 goto exit;
2148 }
2149 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002150 case LAST_BIND:
Al Virofe2d35f2011-03-05 22:58:25 -05002151 /* can't be RCU mode here */
Al Viro16c2cd72011-02-22 15:50:10 -05002152 error = handle_reval_path(nd);
2153 if (error)
2154 goto exit;
Al Viro1f36f772009-12-26 10:56:19 -05002155 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002156 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002157 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002158
Al Viroca344a892011-03-09 00:36:45 -05002159 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002160 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002161 if (nd->last.name[nd->last.len])
2162 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002163 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2164 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002165 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002166 error = walk_component(nd, path, &nd->last, LAST_NORM,
2167 !symlink_ok);
2168 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002169 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002170 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002171 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002172 /* sayonara */
2173 if (nd->flags & LOOKUP_RCU) {
2174 if (nameidata_drop_rcu_last(nd))
2175 return ERR_PTR(-ECHILD);
2176 }
2177
2178 error = -ENOTDIR;
2179 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002180 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002181 goto exit;
2182 }
2183 audit_inode(pathname, nd->path.dentry);
2184 goto ok;
2185 }
2186
2187 /* create side of things */
2188
2189 if (nd->flags & LOOKUP_RCU) {
2190 if (nameidata_drop_rcu_last(nd))
2191 return ERR_PTR(-ECHILD);
2192 }
2193
2194 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002195 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002196 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002197 if (nd->last.name[nd->last.len])
2198 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002199
Al Viroa1e28032009-12-24 02:12:06 -05002200 mutex_lock(&dir->d_inode->i_mutex);
2201
Al Viro6c0d46c2011-03-09 00:59:59 -05002202 dentry = lookup_hash(nd);
2203 error = PTR_ERR(dentry);
2204 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002205 mutex_unlock(&dir->d_inode->i_mutex);
2206 goto exit;
2207 }
2208
Al Viro6c0d46c2011-03-09 00:59:59 -05002209 path->dentry = dentry;
2210 path->mnt = nd->path.mnt;
2211
Al Virofb1cc552009-12-24 01:58:28 -05002212 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002213 if (!dentry->d_inode) {
2214 int mode = op->mode;
2215 if (!IS_POSIXACL(dir->d_inode))
2216 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002217 /*
2218 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002219 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002220 * the time when the file is created and when
2221 * a permanent write count is taken through
2222 * the 'struct file' in nameidata_to_filp().
2223 */
2224 error = mnt_want_write(nd->path.mnt);
2225 if (error)
2226 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002227 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002228 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002229 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002230 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002231 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002232 error = security_path_mknod(&nd->path, dentry, mode, 0);
2233 if (error)
2234 goto exit_mutex_unlock;
2235 error = vfs_create(dir->d_inode, dentry, mode, nd);
2236 if (error)
2237 goto exit_mutex_unlock;
2238 mutex_unlock(&dir->d_inode->i_mutex);
2239 dput(nd->path.dentry);
2240 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002241 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002242 }
2243
2244 /*
2245 * It already exists.
2246 */
2247 mutex_unlock(&dir->d_inode->i_mutex);
2248 audit_inode(pathname, path->dentry);
2249
2250 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002251 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002252 goto exit_dput;
2253
David Howells9875cf82011-01-14 18:45:21 +00002254 error = follow_managed(path, nd->flags);
2255 if (error < 0)
2256 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002257
2258 error = -ENOENT;
2259 if (!path->dentry->d_inode)
2260 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002261
2262 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002263 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002264
2265 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002266 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002267 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002268 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002269 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002270ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002271 if (!S_ISREG(nd->inode->i_mode))
2272 will_truncate = 0;
2273
Al Viro0f9d1a12011-03-09 00:13:14 -05002274 if (will_truncate) {
2275 error = mnt_want_write(nd->path.mnt);
2276 if (error)
2277 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002278 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002279 }
Al Viroca344a892011-03-09 00:36:45 -05002280common:
Al Virobcda7652011-03-13 16:42:14 -04002281 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002282 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002283 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002284 filp = nameidata_to_filp(nd);
2285 if (!IS_ERR(filp)) {
2286 error = ima_file_check(filp, op->acc_mode);
2287 if (error) {
2288 fput(filp);
2289 filp = ERR_PTR(error);
2290 }
2291 }
2292 if (!IS_ERR(filp)) {
2293 if (will_truncate) {
2294 error = handle_truncate(filp);
2295 if (error) {
2296 fput(filp);
2297 filp = ERR_PTR(error);
2298 }
2299 }
2300 }
Al Viroca344a892011-03-09 00:36:45 -05002301out:
2302 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002303 mnt_drop_write(nd->path.mnt);
2304 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002305 return filp;
2306
2307exit_mutex_unlock:
2308 mutex_unlock(&dir->d_inode->i_mutex);
2309exit_dput:
2310 path_put_conditional(path, nd);
2311exit:
Al Viroca344a892011-03-09 00:36:45 -05002312 filp = ERR_PTR(error);
2313 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002314}
2315
Al Viro13aab422011-02-23 17:54:08 -05002316static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002317 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318{
Al Virofe2d35f2011-03-05 22:58:25 -05002319 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002320 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002321 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002322 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002323
2324 filp = get_empty_filp();
2325 if (!filp)
2326 return ERR_PTR(-ENFILE);
2327
Al Viro47c805d2011-02-23 17:44:09 -05002328 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002329 nd->intent.open.file = filp;
2330 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2331 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002332
Al Viro73d049a2011-03-11 12:08:24 -05002333 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002334 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002335 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002336
Al Virofe2d35f2011-03-05 22:58:25 -05002337 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002338 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002339 if (unlikely(error))
2340 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Al Viro73d049a2011-03-11 12:08:24 -05002342 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002343 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002344 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002345 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002346 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002347 path_put_conditional(&path, nd);
2348 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002349 filp = ERR_PTR(-ELOOP);
2350 break;
2351 }
Al Viro73d049a2011-03-11 12:08:24 -05002352 nd->flags |= LOOKUP_PARENT;
2353 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002354 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002355 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002356 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002357 else
Al Viro73d049a2011-03-11 12:08:24 -05002358 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002359 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002360 }
Al Viro10fa8e62009-12-26 07:09:49 -05002361out:
Al Viro73d049a2011-03-11 12:08:24 -05002362 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2363 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002364 if (base)
2365 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002366 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002367 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
Nick Piggin31e6b012011-01-07 17:49:52 +11002369out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002370 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002371 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372}
2373
Al Viro13aab422011-02-23 17:54:08 -05002374struct file *do_filp_open(int dfd, const char *pathname,
2375 const struct open_flags *op, int flags)
2376{
Al Viro73d049a2011-03-11 12:08:24 -05002377 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002378 struct file *filp;
2379
Al Viro73d049a2011-03-11 12:08:24 -05002380 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002381 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002382 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002383 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002384 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002385 return filp;
2386}
2387
Al Viro73d049a2011-03-11 12:08:24 -05002388struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2389 const char *name, const struct open_flags *op, int flags)
2390{
2391 struct nameidata nd;
2392 struct file *file;
2393
2394 nd.root.mnt = mnt;
2395 nd.root.dentry = dentry;
2396
2397 flags |= LOOKUP_ROOT;
2398
Al Virobcda7652011-03-13 16:42:14 -04002399 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002400 return ERR_PTR(-ELOOP);
2401
2402 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2403 if (unlikely(file == ERR_PTR(-ECHILD)))
2404 file = path_openat(-1, name, &nd, op, flags);
2405 if (unlikely(file == ERR_PTR(-ESTALE)))
2406 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2407 return file;
2408}
2409
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410/**
2411 * lookup_create - lookup a dentry, creating it if it doesn't exist
2412 * @nd: nameidata info
2413 * @is_dir: directory flag
2414 *
2415 * Simple function to lookup and return a dentry and create it
2416 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002417 *
Jan Blunck4ac91372008-02-14 19:34:32 -08002418 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 */
2420struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2421{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002422 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Jan Blunck4ac91372008-02-14 19:34:32 -08002424 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002425 /*
2426 * Yucky last component or no last component at all?
2427 * (foo/., foo/.., /////)
2428 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 if (nd->last_type != LAST_NORM)
2430 goto fail;
2431 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04002432 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04002433 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002434
2435 /*
2436 * Do the final lookup.
2437 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08002438 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 if (IS_ERR(dentry))
2440 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002441
Al Viroe9baf6e2008-05-15 04:49:12 -04002442 if (dentry->d_inode)
2443 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002444 /*
2445 * Special case - lookup gave negative, but... we had foo/bar/
2446 * From the vfs_mknod() POV we just have a negative dentry -
2447 * all is fine. Let's be bastards - you had / on the end, you've
2448 * been asking for (non-existent) directory. -ENOENT for you.
2449 */
Al Viroe9baf6e2008-05-15 04:49:12 -04002450 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2451 dput(dentry);
2452 dentry = ERR_PTR(-ENOENT);
2453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002455eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002457 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458fail:
2459 return dentry;
2460}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07002461EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462
2463int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2464{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002465 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
2467 if (error)
2468 return error;
2469
Serge E. Hallyne795b712011-03-23 16:43:25 -07002470 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2471 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 return -EPERM;
2473
Al Viroacfa4382008-12-04 10:06:33 -05002474 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 return -EPERM;
2476
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002477 error = devcgroup_inode_mknod(mode, dev);
2478 if (error)
2479 return error;
2480
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 error = security_inode_mknod(dir, dentry, mode, dev);
2482 if (error)
2483 return error;
2484
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002486 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002487 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 return error;
2489}
2490
Dave Hansen463c3192008-02-15 14:37:57 -08002491static int may_mknod(mode_t mode)
2492{
2493 switch (mode & S_IFMT) {
2494 case S_IFREG:
2495 case S_IFCHR:
2496 case S_IFBLK:
2497 case S_IFIFO:
2498 case S_IFSOCK:
2499 case 0: /* zero mode translates to S_IFREG */
2500 return 0;
2501 case S_IFDIR:
2502 return -EPERM;
2503 default:
2504 return -EINVAL;
2505 }
2506}
2507
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002508SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2509 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510{
Al Viro2ad94ae2008-07-21 09:32:51 -04002511 int error;
2512 char *tmp;
2513 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 struct nameidata nd;
2515
2516 if (S_ISDIR(mode))
2517 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
Al Viro2ad94ae2008-07-21 09:32:51 -04002519 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002521 return error;
2522
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002524 if (IS_ERR(dentry)) {
2525 error = PTR_ERR(dentry);
2526 goto out_unlock;
2527 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002528 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002529 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002530 error = may_mknod(mode);
2531 if (error)
2532 goto out_dput;
2533 error = mnt_want_write(nd.path.mnt);
2534 if (error)
2535 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002536 error = security_path_mknod(&nd.path, dentry, mode, dev);
2537 if (error)
2538 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002539 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002541 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 break;
2543 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002544 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 new_decode_dev(dev));
2546 break;
2547 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002548 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002551out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002552 mnt_drop_write(nd.path.mnt);
2553out_dput:
2554 dput(dentry);
2555out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002556 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002557 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 putname(tmp);
2559
2560 return error;
2561}
2562
Heiko Carstens3480b252009-01-14 14:14:16 +01002563SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002564{
2565 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2566}
2567
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2569{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002570 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
2572 if (error)
2573 return error;
2574
Al Viroacfa4382008-12-04 10:06:33 -05002575 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 return -EPERM;
2577
2578 mode &= (S_IRWXUGO|S_ISVTX);
2579 error = security_inode_mkdir(dir, dentry, mode);
2580 if (error)
2581 return error;
2582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002584 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002585 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 return error;
2587}
2588
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002589SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590{
2591 int error = 0;
2592 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002593 struct dentry *dentry;
2594 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
Al Viro2ad94ae2008-07-21 09:32:51 -04002596 error = user_path_parent(dfd, pathname, &nd, &tmp);
2597 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002598 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599
Dave Hansen6902d922006-09-30 23:29:01 -07002600 dentry = lookup_create(&nd, 1);
2601 error = PTR_ERR(dentry);
2602 if (IS_ERR(dentry))
2603 goto out_unlock;
2604
Jan Blunck4ac91372008-02-14 19:34:32 -08002605 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002606 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002607 error = mnt_want_write(nd.path.mnt);
2608 if (error)
2609 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002610 error = security_path_mkdir(&nd.path, dentry, mode);
2611 if (error)
2612 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002613 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002614out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002615 mnt_drop_write(nd.path.mnt);
2616out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002617 dput(dentry);
2618out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002619 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002620 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002621 putname(tmp);
2622out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 return error;
2624}
2625
Heiko Carstens3cdad422009-01-14 14:14:22 +01002626SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002627{
2628 return sys_mkdirat(AT_FDCWD, pathname, mode);
2629}
2630
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631/*
2632 * We try to drop the dentry early: we should have
2633 * a usage count of 2 if we're the only user of this
2634 * dentry, and if that is true (possibly after pruning
2635 * the dcache), then we drop the dentry now.
2636 *
2637 * A low-level filesystem can, if it choses, legally
2638 * do a
2639 *
2640 * if (!d_unhashed(dentry))
2641 * return -EBUSY;
2642 *
2643 * if it cannot handle the case of removing a directory
2644 * that is still in use by something else..
2645 */
2646void dentry_unhash(struct dentry *dentry)
2647{
2648 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002649 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002651 if (dentry->d_count == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 __d_drop(dentry);
2653 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654}
2655
2656int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2657{
2658 int error = may_delete(dir, dentry, 1);
2659
2660 if (error)
2661 return error;
2662
Al Viroacfa4382008-12-04 10:06:33 -05002663 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 return -EPERM;
2665
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002666 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 dentry_unhash(dentry);
2668 if (d_mountpoint(dentry))
2669 error = -EBUSY;
2670 else {
2671 error = security_inode_rmdir(dir, dentry);
2672 if (!error) {
2673 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002674 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002676 dont_mount(dentry);
2677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 }
2679 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002680 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 d_delete(dentry);
2683 }
2684 dput(dentry);
2685
2686 return error;
2687}
2688
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002689static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690{
2691 int error = 0;
2692 char * name;
2693 struct dentry *dentry;
2694 struct nameidata nd;
2695
Al Viro2ad94ae2008-07-21 09:32:51 -04002696 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002698 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
2700 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002701 case LAST_DOTDOT:
2702 error = -ENOTEMPTY;
2703 goto exit1;
2704 case LAST_DOT:
2705 error = -EINVAL;
2706 goto exit1;
2707 case LAST_ROOT:
2708 error = -EBUSY;
2709 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002711
2712 nd.flags &= ~LOOKUP_PARENT;
2713
Jan Blunck4ac91372008-02-14 19:34:32 -08002714 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002715 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002717 if (IS_ERR(dentry))
2718 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002719 error = mnt_want_write(nd.path.mnt);
2720 if (error)
2721 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002722 error = security_path_rmdir(&nd.path, dentry);
2723 if (error)
2724 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002725 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002726exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002727 mnt_drop_write(nd.path.mnt);
2728exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002729 dput(dentry);
2730exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002731 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002733 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 putname(name);
2735 return error;
2736}
2737
Heiko Carstens3cdad422009-01-14 14:14:22 +01002738SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002739{
2740 return do_rmdir(AT_FDCWD, pathname);
2741}
2742
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743int vfs_unlink(struct inode *dir, struct dentry *dentry)
2744{
2745 int error = may_delete(dir, dentry, 0);
2746
2747 if (error)
2748 return error;
2749
Al Viroacfa4382008-12-04 10:06:33 -05002750 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 return -EPERM;
2752
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002753 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 if (d_mountpoint(dentry))
2755 error = -EBUSY;
2756 else {
2757 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002758 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002760 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002761 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002764 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
2766 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2767 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002768 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 }
Robert Love0eeca282005-07-12 17:06:03 -04002771
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 return error;
2773}
2774
2775/*
2776 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002777 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 * writeout happening, and we don't want to prevent access to the directory
2779 * while waiting on the I/O.
2780 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002781static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782{
Al Viro2ad94ae2008-07-21 09:32:51 -04002783 int error;
2784 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 struct dentry *dentry;
2786 struct nameidata nd;
2787 struct inode *inode = NULL;
2788
Al Viro2ad94ae2008-07-21 09:32:51 -04002789 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002791 return error;
2792
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 error = -EISDIR;
2794 if (nd.last_type != LAST_NORM)
2795 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002796
2797 nd.flags &= ~LOOKUP_PARENT;
2798
Jan Blunck4ac91372008-02-14 19:34:32 -08002799 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002800 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 error = PTR_ERR(dentry);
2802 if (!IS_ERR(dentry)) {
2803 /* Why not before? Because we want correct error value */
2804 if (nd.last.name[nd.last.len])
2805 goto slashes;
2806 inode = dentry->d_inode;
2807 if (inode)
Al Viro7de9c6ee2010-10-23 11:11:40 -04002808 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002809 error = mnt_want_write(nd.path.mnt);
2810 if (error)
2811 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002812 error = security_path_unlink(&nd.path, dentry);
2813 if (error)
2814 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002815 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002816exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002817 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 exit2:
2819 dput(dentry);
2820 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002821 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 if (inode)
2823 iput(inode); /* truncate the inode here */
2824exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002825 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 putname(name);
2827 return error;
2828
2829slashes:
2830 error = !dentry->d_inode ? -ENOENT :
2831 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2832 goto exit2;
2833}
2834
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002835SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002836{
2837 if ((flag & ~AT_REMOVEDIR) != 0)
2838 return -EINVAL;
2839
2840 if (flag & AT_REMOVEDIR)
2841 return do_rmdir(dfd, pathname);
2842
2843 return do_unlinkat(dfd, pathname);
2844}
2845
Heiko Carstens3480b252009-01-14 14:14:16 +01002846SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002847{
2848 return do_unlinkat(AT_FDCWD, pathname);
2849}
2850
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002851int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002853 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
2855 if (error)
2856 return error;
2857
Al Viroacfa4382008-12-04 10:06:33 -05002858 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 return -EPERM;
2860
2861 error = security_inode_symlink(dir, dentry, oldname);
2862 if (error)
2863 return error;
2864
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002866 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002867 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 return error;
2869}
2870
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002871SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2872 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873{
Al Viro2ad94ae2008-07-21 09:32:51 -04002874 int error;
2875 char *from;
2876 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002877 struct dentry *dentry;
2878 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
2880 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002881 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002883
2884 error = user_path_parent(newdfd, newname, &nd, &to);
2885 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002886 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887
Dave Hansen6902d922006-09-30 23:29:01 -07002888 dentry = lookup_create(&nd, 0);
2889 error = PTR_ERR(dentry);
2890 if (IS_ERR(dentry))
2891 goto out_unlock;
2892
Dave Hansen75c3f292008-02-15 14:37:45 -08002893 error = mnt_want_write(nd.path.mnt);
2894 if (error)
2895 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002896 error = security_path_symlink(&nd.path, dentry, from);
2897 if (error)
2898 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002899 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002900out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002901 mnt_drop_write(nd.path.mnt);
2902out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002903 dput(dentry);
2904out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002905 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002906 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002907 putname(to);
2908out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 putname(from);
2910 return error;
2911}
2912
Heiko Carstens3480b252009-01-14 14:14:16 +01002913SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002914{
2915 return sys_symlinkat(oldname, AT_FDCWD, newname);
2916}
2917
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2919{
2920 struct inode *inode = old_dentry->d_inode;
2921 int error;
2922
2923 if (!inode)
2924 return -ENOENT;
2925
Miklos Szeredia95164d2008-07-30 15:08:48 +02002926 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 if (error)
2928 return error;
2929
2930 if (dir->i_sb != inode->i_sb)
2931 return -EXDEV;
2932
2933 /*
2934 * A link to an append-only or immutable file cannot be created.
2935 */
2936 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2937 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002938 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002940 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 return -EPERM;
2942
2943 error = security_inode_link(old_dentry, dir, new_dentry);
2944 if (error)
2945 return error;
2946
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002947 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302948 /* Make sure we don't allow creating hardlink to an unlinked file */
2949 if (inode->i_nlink == 0)
2950 error = -ENOENT;
2951 else
2952 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002953 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002954 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002955 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 return error;
2957}
2958
2959/*
2960 * Hardlinks are often used in delicate situations. We avoid
2961 * security-related surprises by not following symlinks on the
2962 * newname. --KAB
2963 *
2964 * We don't follow them on the oldname either to be compatible
2965 * with linux 2.0, and to avoid hard-linking to directories
2966 * and other special files. --ADM
2967 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002968SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2969 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970{
2971 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002972 struct nameidata nd;
2973 struct path old_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302974 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002976 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302978 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002979 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302980 /*
2981 * To use null names we require CAP_DAC_READ_SEARCH
2982 * This ensures that not everyone will be able to create
2983 * handlink using the passed filedescriptor.
2984 */
2985 if (flags & AT_EMPTY_PATH) {
2986 if (!capable(CAP_DAC_READ_SEARCH))
2987 return -ENOENT;
2988 how = LOOKUP_EMPTY;
2989 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002990
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302991 if (flags & AT_SYMLINK_FOLLOW)
2992 how |= LOOKUP_FOLLOW;
2993
2994 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002996 return error;
2997
2998 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 if (error)
3000 goto out;
3001 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04003002 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 goto out_release;
3004 new_dentry = lookup_create(&nd, 0);
3005 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003006 if (IS_ERR(new_dentry))
3007 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08003008 error = mnt_want_write(nd.path.mnt);
3009 if (error)
3010 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003011 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3012 if (error)
3013 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04003014 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003015out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08003016 mnt_drop_write(nd.path.mnt);
3017out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003018 dput(new_dentry);
3019out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08003020 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08003022 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003023 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024out:
Al Viro2d8f3032008-07-22 09:59:21 -04003025 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026
3027 return error;
3028}
3029
Heiko Carstens3480b252009-01-14 14:14:16 +01003030SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003031{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003032 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003033}
3034
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035/*
3036 * The worst of all namespace operations - renaming directory. "Perverted"
3037 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3038 * Problems:
3039 * a) we can get into loop creation. Check is done in is_subdir().
3040 * b) race potential - two innocent renames can create a loop together.
3041 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003042 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 * story.
3044 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003045 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 * whether the target exists). Solution: try to be smart with locking
3047 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003048 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 * move will be locked. Thus we can rank directories by the tree
3050 * (ancestors first) and rank all non-directories after them.
3051 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003052 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 * HOWEVER, it relies on the assumption that any object with ->lookup()
3054 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3055 * we'd better make sure that there's no link(2) for them.
3056 * d) some filesystems don't support opened-but-unlinked directories,
3057 * either because of layout or because they are not ready to deal with
3058 * all cases correctly. The latter will be fixed (taking this sort of
3059 * stuff into VFS), but the former is not going away. Solution: the same
3060 * trick as in rmdir().
3061 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003062 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003064 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 * locking].
3066 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003067static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3068 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069{
3070 int error = 0;
3071 struct inode *target;
3072
3073 /*
3074 * If we are going to change the parent - check write permissions,
3075 * we'll need to flip '..'.
3076 */
3077 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003078 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 if (error)
3080 return error;
3081 }
3082
3083 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3084 if (error)
3085 return error;
3086
3087 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04003088 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003089 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3091 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04003092 else {
3093 if (target)
3094 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04003098 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04003100 dont_mount(new_dentry);
3101 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003102 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 if (d_unhashed(new_dentry))
3104 d_rehash(new_dentry);
3105 dput(new_dentry);
3106 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003107 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003108 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3109 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110 return error;
3111}
3112
Adrian Bunk75c96f82005-05-05 16:16:09 -07003113static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3114 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115{
3116 struct inode *target;
3117 int error;
3118
3119 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3120 if (error)
3121 return error;
3122
3123 dget(new_dentry);
3124 target = new_dentry->d_inode;
3125 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003126 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3128 error = -EBUSY;
3129 else
3130 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3131 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05003132 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04003133 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07003134 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 }
3137 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003138 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 dput(new_dentry);
3140 return error;
3141}
3142
3143int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3144 struct inode *new_dir, struct dentry *new_dentry)
3145{
3146 int error;
3147 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003148 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149
3150 if (old_dentry->d_inode == new_dentry->d_inode)
3151 return 0;
3152
3153 error = may_delete(old_dir, old_dentry, is_dir);
3154 if (error)
3155 return error;
3156
3157 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003158 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 else
3160 error = may_delete(new_dir, new_dentry, is_dir);
3161 if (error)
3162 return error;
3163
Al Viroacfa4382008-12-04 10:06:33 -05003164 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 return -EPERM;
3166
Robert Love0eeca282005-07-12 17:06:03 -04003167 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3168
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 if (is_dir)
3170 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3171 else
3172 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003173 if (!error)
3174 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003175 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003176 fsnotify_oldname_free(old_name);
3177
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 return error;
3179}
3180
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003181SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3182 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183{
Al Viro2ad94ae2008-07-21 09:32:51 -04003184 struct dentry *old_dir, *new_dir;
3185 struct dentry *old_dentry, *new_dentry;
3186 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003188 char *from;
3189 char *to;
3190 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
Al Viro2ad94ae2008-07-21 09:32:51 -04003192 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 if (error)
3194 goto exit;
3195
Al Viro2ad94ae2008-07-21 09:32:51 -04003196 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 if (error)
3198 goto exit1;
3199
3200 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003201 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202 goto exit2;
3203
Jan Blunck4ac91372008-02-14 19:34:32 -08003204 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 error = -EBUSY;
3206 if (oldnd.last_type != LAST_NORM)
3207 goto exit2;
3208
Jan Blunck4ac91372008-02-14 19:34:32 -08003209 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 if (newnd.last_type != LAST_NORM)
3211 goto exit2;
3212
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003213 oldnd.flags &= ~LOOKUP_PARENT;
3214 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003215 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003216
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217 trap = lock_rename(new_dir, old_dir);
3218
Christoph Hellwig49705b72005-11-08 21:35:06 -08003219 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 error = PTR_ERR(old_dentry);
3221 if (IS_ERR(old_dentry))
3222 goto exit3;
3223 /* source must exist */
3224 error = -ENOENT;
3225 if (!old_dentry->d_inode)
3226 goto exit4;
3227 /* unless the source is a directory trailing slashes give -ENOTDIR */
3228 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3229 error = -ENOTDIR;
3230 if (oldnd.last.name[oldnd.last.len])
3231 goto exit4;
3232 if (newnd.last.name[newnd.last.len])
3233 goto exit4;
3234 }
3235 /* source should not be ancestor of target */
3236 error = -EINVAL;
3237 if (old_dentry == trap)
3238 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003239 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 error = PTR_ERR(new_dentry);
3241 if (IS_ERR(new_dentry))
3242 goto exit4;
3243 /* target should not be an ancestor of source */
3244 error = -ENOTEMPTY;
3245 if (new_dentry == trap)
3246 goto exit5;
3247
Dave Hansen9079b1e2008-02-15 14:37:49 -08003248 error = mnt_want_write(oldnd.path.mnt);
3249 if (error)
3250 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003251 error = security_path_rename(&oldnd.path, old_dentry,
3252 &newnd.path, new_dentry);
3253 if (error)
3254 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 error = vfs_rename(old_dir->d_inode, old_dentry,
3256 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003257exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003258 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259exit5:
3260 dput(new_dentry);
3261exit4:
3262 dput(old_dentry);
3263exit3:
3264 unlock_rename(new_dir, old_dir);
3265exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003266 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003267 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003269 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003271exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 return error;
3273}
3274
Heiko Carstensa26eab22009-01-14 14:14:17 +01003275SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003276{
3277 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3278}
3279
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3281{
3282 int len;
3283
3284 len = PTR_ERR(link);
3285 if (IS_ERR(link))
3286 goto out;
3287
3288 len = strlen(link);
3289 if (len > (unsigned) buflen)
3290 len = buflen;
3291 if (copy_to_user(buffer, link, len))
3292 len = -EFAULT;
3293out:
3294 return len;
3295}
3296
3297/*
3298 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3299 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3300 * using) it for any given inode is up to filesystem.
3301 */
3302int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3303{
3304 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003305 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003306 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003307
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003309 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003310 if (IS_ERR(cookie))
3311 return PTR_ERR(cookie);
3312
3313 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3314 if (dentry->d_inode->i_op->put_link)
3315 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3316 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317}
3318
3319int vfs_follow_link(struct nameidata *nd, const char *link)
3320{
3321 return __vfs_follow_link(nd, link);
3322}
3323
3324/* get the link contents into pagecache */
3325static char *page_getlink(struct dentry * dentry, struct page **ppage)
3326{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003327 char *kaddr;
3328 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003330 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003332 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003334 kaddr = kmap(page);
3335 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3336 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337}
3338
3339int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3340{
3341 struct page *page = NULL;
3342 char *s = page_getlink(dentry, &page);
3343 int res = vfs_readlink(dentry,buffer,buflen,s);
3344 if (page) {
3345 kunmap(page);
3346 page_cache_release(page);
3347 }
3348 return res;
3349}
3350
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003351void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003353 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003355 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356}
3357
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003358void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003360 struct page *page = cookie;
3361
3362 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 kunmap(page);
3364 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 }
3366}
3367
Nick Piggin54566b22009-01-04 12:00:53 -08003368/*
3369 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3370 */
3371int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372{
3373 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003374 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003375 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003376 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003378 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3379 if (nofs)
3380 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
NeilBrown7e53cac2006-03-25 03:07:57 -08003382retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003383 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003384 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003386 goto fail;
3387
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 kaddr = kmap_atomic(page, KM_USER0);
3389 memcpy(kaddr, symname, len-1);
3390 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003391
3392 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3393 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 if (err < 0)
3395 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003396 if (err < len-1)
3397 goto retry;
3398
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 mark_inode_dirty(inode);
3400 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401fail:
3402 return err;
3403}
3404
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003405int page_symlink(struct inode *inode, const char *symname, int len)
3406{
3407 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003408 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003409}
3410
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003411const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412 .readlink = generic_readlink,
3413 .follow_link = page_follow_link_light,
3414 .put_link = page_put_link,
3415};
3416
Al Viro2d8f3032008-07-22 09:59:21 -04003417EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003418EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419EXPORT_SYMBOL(follow_down);
3420EXPORT_SYMBOL(follow_up);
3421EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3422EXPORT_SYMBOL(getname);
3423EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424EXPORT_SYMBOL(lookup_one_len);
3425EXPORT_SYMBOL(page_follow_link_light);
3426EXPORT_SYMBOL(page_put_link);
3427EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003428EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429EXPORT_SYMBOL(page_symlink);
3430EXPORT_SYMBOL(page_symlink_inode_operations);
Al Viroc9c6cac2011-02-16 15:15:47 -05003431EXPORT_SYMBOL(kern_path_parent);
Al Virod1811462008-08-02 00:49:18 -04003432EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003433EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003434EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08003435EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436EXPORT_SYMBOL(unlock_rename);
3437EXPORT_SYMBOL(vfs_create);
3438EXPORT_SYMBOL(vfs_follow_link);
3439EXPORT_SYMBOL(vfs_link);
3440EXPORT_SYMBOL(vfs_mkdir);
3441EXPORT_SYMBOL(vfs_mknod);
3442EXPORT_SYMBOL(generic_permission);
3443EXPORT_SYMBOL(vfs_readlink);
3444EXPORT_SYMBOL(vfs_rename);
3445EXPORT_SYMBOL(vfs_rmdir);
3446EXPORT_SYMBOL(vfs_symlink);
3447EXPORT_SYMBOL(vfs_unlink);
3448EXPORT_SYMBOL(dentry_unhash);
3449EXPORT_SYMBOL(generic_readlink);