blob: 1ab641f2e78ebdddff0306073ddbb0d06685f004 [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 Blunck5dd784d2008-02-14 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
Al Viro19660af2011-03-25 10:32:48 -0400394/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100395 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400396 * Documentation/filesystems/path-lookup.txt). In situations when we can't
397 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
398 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
399 * mode. Refcounts are grabbed at the last known good point before rcu-walk
400 * got stuck, so ref-walk may continue from there. If this is not successful
401 * (eg. a seqcount has changed), then failure is returned and it's up to caller
402 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100403 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100404
405/**
Al Viro19660af2011-03-25 10:32:48 -0400406 * unlazy_walk - try to switch to ref-walk mode.
407 * @nd: nameidata pathwalk data
408 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800409 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100410 *
Al Viro19660af2011-03-25 10:32:48 -0400411 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
412 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
413 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100414 */
Al Viro19660af2011-03-25 10:32:48 -0400415static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100416{
417 struct fs_struct *fs = current->fs;
418 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500419 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100420
421 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500422 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
423 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100424 spin_lock(&fs->lock);
425 if (nd->root.mnt != fs->root.mnt ||
426 nd->root.dentry != fs->root.dentry)
427 goto err_root;
428 }
429 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400430 if (!dentry) {
431 if (!__d_rcu_to_refcount(parent, nd->seq))
432 goto err_parent;
433 BUG_ON(nd->inode != parent->d_inode);
434 } else {
435 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
436 if (!__d_rcu_to_refcount(dentry, nd->seq))
437 goto err_child;
438 /*
439 * If the sequence check on the child dentry passed, then
440 * the child has not been removed from its parent. This
441 * means the parent dentry must be valid and able to take
442 * a reference at this point.
443 */
444 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
445 BUG_ON(!parent->d_count);
446 parent->d_count++;
447 spin_unlock(&dentry->d_lock);
448 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100449 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500450 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100451 path_get(&nd->root);
452 spin_unlock(&fs->lock);
453 }
454 mntget(nd->path.mnt);
455
456 rcu_read_unlock();
457 br_read_unlock(vfsmount_lock);
458 nd->flags &= ~LOOKUP_RCU;
459 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400460
461err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100462 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400463err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100464 spin_unlock(&parent->d_lock);
465err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500466 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100467 spin_unlock(&fs->lock);
468 return -ECHILD;
469}
470
Nick Piggin31e6b012011-01-07 17:49:52 +1100471/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700472 * release_open_intent - free up open intent resources
473 * @nd: pointer to nameidata
474 */
475void release_open_intent(struct nameidata *nd)
476{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800477 struct file *file = nd->intent.open.file;
478
479 if (file && !IS_ERR(file)) {
480 if (file->f_path.dentry == NULL)
481 put_filp(file);
482 else
483 fput(file);
484 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700485}
486
Al Virof60aef72011-02-15 01:35:28 -0500487static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100488{
Al Virof60aef72011-02-15 01:35:28 -0500489 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100490}
491
Al Virof5e1c1c2011-02-15 01:32:55 -0500492static struct dentry *
Ian Kentbcdc5e02006-09-27 01:50:44 -0700493do_revalidate(struct dentry *dentry, struct nameidata *nd)
494{
Al Virof5e1c1c2011-02-15 01:32:55 -0500495 int status = d_revalidate(dentry, nd);
Ian Kentbcdc5e02006-09-27 01:50:44 -0700496 if (unlikely(status <= 0)) {
497 /*
498 * The dentry failed validation.
499 * If d_revalidate returned 0 attempt to invalidate
500 * the dentry otherwise d_revalidate is asking us
501 * to return a fail status.
502 */
Nick Piggin34286d62011-01-07 17:49:57 +1100503 if (status < 0) {
Al Virof5e1c1c2011-02-15 01:32:55 -0500504 dput(dentry);
Nick Piggin34286d62011-01-07 17:49:57 +1100505 dentry = ERR_PTR(status);
Al Virof5e1c1c2011-02-15 01:32:55 -0500506 } else if (!d_invalidate(dentry)) {
507 dput(dentry);
508 dentry = NULL;
Ian Kentbcdc5e02006-09-27 01:50:44 -0700509 }
510 }
511 return dentry;
512}
513
Al Viro9f1fafe2011-03-25 11:00:12 -0400514/**
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500517 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500523 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400524static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500525{
Al Viro16c2cd72011-02-22 15:50:10 -0500526 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500527 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500528
Al Viro9f1fafe2011-03-25 11:00:12 -0400529 if (nd->flags & LOOKUP_RCU) {
530 nd->flags &= ~LOOKUP_RCU;
531 if (!(nd->flags & LOOKUP_ROOT))
532 nd->root.mnt = NULL;
533 spin_lock(&dentry->d_lock);
534 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
535 spin_unlock(&dentry->d_lock);
536 rcu_read_unlock();
537 br_read_unlock(vfsmount_lock);
538 return -ECHILD;
539 }
540 BUG_ON(nd->inode != dentry->d_inode);
541 spin_unlock(&dentry->d_lock);
542 mntget(nd->path.mnt);
543 rcu_read_unlock();
544 br_read_unlock(vfsmount_lock);
545 }
546
Al Viro16c2cd72011-02-22 15:50:10 -0500547 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500548 return 0;
549
Al Viro16c2cd72011-02-22 15:50:10 -0500550 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
551 return 0;
552
553 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
554 return 0;
555
556 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100557 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500558 if (status > 0)
559 return 0;
560
Al Viro16c2cd72011-02-22 15:50:10 -0500561 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500562 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500563
Al Viro9f1fafe2011-03-25 11:00:12 -0400564 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500565 return status;
566}
567
568/*
Al Virob75b5082009-12-16 01:01:38 -0500569 * Short-cut version of permission(), for calling on directories
570 * during pathname resolution. Combines parts of permission()
571 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 *
573 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500574 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 * complete permission check.
576 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100577static inline int exec_permission(struct inode *inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700579 int ret;
Serge E. Hallyne795b712011-03-23 16:43:25 -0700580 struct user_namespace *ns = inode_userns(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700582 if (inode->i_op->permission) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100583 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
584 } else {
585 ret = acl_permission_check(inode, MAY_EXEC, flags,
586 inode->i_op->check_acl);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700587 }
Nick Pigginb74c79e2011-01-07 17:49:58 +1100588 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 goto ok;
Nick Pigginb74c79e2011-01-07 17:49:58 +1100590 if (ret == -ECHILD)
Nick Piggin31e6b012011-01-07 17:49:52 +1100591 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Serge E. Hallyne795b712011-03-23 16:43:25 -0700593 if (ns_capable(ns, CAP_DAC_OVERRIDE) ||
594 ns_capable(ns, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 goto ok;
596
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700597 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598ok:
Nick Pigginb74c79e2011-01-07 17:49:58 +1100599 return security_inode_exec_permission(inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
Al Viro2a737872009-04-07 11:49:53 -0400602static __always_inline void set_root(struct nameidata *nd)
603{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200604 if (!nd->root.mnt)
605 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400606}
607
Al Viro6de88d72009-08-09 01:41:57 +0400608static int link_path_walk(const char *, struct nameidata *);
609
Nick Piggin31e6b012011-01-07 17:49:52 +1100610static __always_inline void set_root_rcu(struct nameidata *nd)
611{
612 if (!nd->root.mnt) {
613 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100614 unsigned seq;
615
616 do {
617 seq = read_seqcount_begin(&fs->seq);
618 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700619 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100620 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100621 }
622}
623
Arjan van de Venf1662352006-01-14 13:21:31 -0800624static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
Nick Piggin31e6b012011-01-07 17:49:52 +1100626 int ret;
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (IS_ERR(link))
629 goto fail;
630
631 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400632 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800633 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400634 nd->path = nd->root;
635 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500636 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100638 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100639
Nick Piggin31e6b012011-01-07 17:49:52 +1100640 ret = link_path_walk(link, nd);
641 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800643 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 return PTR_ERR(link);
645}
646
Jan Blunck1d957f92008-02-14 19:34:35 -0800647static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700648{
649 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800650 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700651 mntput(path->mnt);
652}
653
Nick Piggin7b9337a2011-01-14 08:42:43 +0000654static inline void path_to_nameidata(const struct path *path,
655 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700656{
Nick Piggin31e6b012011-01-07 17:49:52 +1100657 if (!(nd->flags & LOOKUP_RCU)) {
658 dput(nd->path.dentry);
659 if (nd->path.mnt != path->mnt)
660 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800661 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100662 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800663 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700664}
665
Al Viro574197e2011-03-14 22:20:34 -0400666static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
667{
668 struct inode *inode = link->dentry->d_inode;
669 if (!IS_ERR(cookie) && inode->i_op->put_link)
670 inode->i_op->put_link(link->dentry, nd, cookie);
671 path_put(link);
672}
673
Al Virodef4af32009-12-26 08:37:05 -0500674static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400675follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800676{
677 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000678 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800679
Al Viro844a3912011-02-15 00:38:26 -0500680 BUG_ON(nd->flags & LOOKUP_RCU);
681
Al Viro0e794582011-03-16 02:45:02 -0400682 if (link->mnt == nd->path.mnt)
683 mntget(link->mnt);
684
Al Viro574197e2011-03-14 22:20:34 -0400685 if (unlikely(current->total_link_count >= 40)) {
686 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400687 path_put(&nd->path);
688 return -ELOOP;
689 }
690 cond_resched();
691 current->total_link_count++;
692
Nick Piggin7b9337a2011-01-14 08:42:43 +0000693 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800694 nd_set_link(nd, NULL);
695
Al Viro36f3b4f2011-02-22 21:24:38 -0500696 error = security_inode_follow_link(link->dentry, nd);
697 if (error) {
698 *p = ERR_PTR(error); /* no ->put_link(), please */
699 path_put(&nd->path);
700 return error;
701 }
702
Al Viro86acdca12009-12-22 23:45:11 -0500703 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500704 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
705 error = PTR_ERR(*p);
706 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800707 char *s = nd_get_link(nd);
708 error = 0;
709 if (s)
710 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400711 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500712 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400713 nd->inode = nd->path.dentry->d_inode;
714 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400715 /* stepped on a _really_ weird one */
716 path_put(&nd->path);
717 error = -ELOOP;
718 }
719 }
Ian Kent051d3812006-03-27 01:14:53 -0800720 }
Ian Kent051d3812006-03-27 01:14:53 -0800721 return error;
722}
723
Nick Piggin31e6b012011-01-07 17:49:52 +1100724static int follow_up_rcu(struct path *path)
725{
726 struct vfsmount *parent;
727 struct dentry *mountpoint;
728
729 parent = path->mnt->mnt_parent;
730 if (parent == path->mnt)
731 return 0;
732 mountpoint = path->mnt->mnt_mountpoint;
733 path->dentry = mountpoint;
734 path->mnt = parent;
735 return 1;
736}
737
Al Virobab77eb2009-04-18 03:26:48 -0400738int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
740 struct vfsmount *parent;
741 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000742
743 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400744 parent = path->mnt->mnt_parent;
745 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000746 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return 0;
748 }
749 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400750 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000751 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400752 dput(path->dentry);
753 path->dentry = mountpoint;
754 mntput(path->mnt);
755 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 return 1;
757}
758
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100759/*
David Howells9875cf82011-01-14 18:45:21 +0000760 * Perform an automount
761 * - return -EISDIR to tell follow_managed() to stop and return the path we
762 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 */
David Howells9875cf82011-01-14 18:45:21 +0000764static int follow_automount(struct path *path, unsigned flags,
765 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100766{
David Howells9875cf82011-01-14 18:45:21 +0000767 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000768 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100769
David Howells9875cf82011-01-14 18:45:21 +0000770 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
771 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700772
David Howells6f45b652011-01-14 18:45:31 +0000773 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
774 * and this is the terminal part of the path.
775 */
776 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
777 return -EISDIR; /* we actually want to stop here */
778
David Howells9875cf82011-01-14 18:45:21 +0000779 /* We want to mount if someone is trying to open/create a file of any
780 * type under the mountpoint, wants to traverse through the mountpoint
781 * or wants to open the mounted directory.
782 *
783 * We don't want to mount if someone's just doing a stat and they've
784 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
785 * appended a '/' to the name.
786 */
787 if (!(flags & LOOKUP_FOLLOW) &&
788 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
789 LOOKUP_OPEN | LOOKUP_CREATE)))
790 return -EISDIR;
791
792 current->total_link_count++;
793 if (current->total_link_count >= 40)
794 return -ELOOP;
795
796 mnt = path->dentry->d_op->d_automount(path);
797 if (IS_ERR(mnt)) {
798 /*
799 * The filesystem is allowed to return -EISDIR here to indicate
800 * it doesn't want to automount. For instance, autofs would do
801 * this so that its userspace daemon can mount on this dentry.
802 *
803 * However, we can only permit this if it's a terminal point in
804 * the path being looked up; if it wasn't then the remainder of
805 * the path is inaccessible and we should say so.
806 */
807 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
808 return -EREMOTE;
809 return PTR_ERR(mnt);
810 }
David Howellsea5b7782011-01-14 19:10:03 +0000811
David Howells9875cf82011-01-14 18:45:21 +0000812 if (!mnt) /* mount collision */
813 return 0;
814
Al Viro19a167a2011-01-17 01:35:23 -0500815 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000816
David Howellsea5b7782011-01-14 19:10:03 +0000817 switch (err) {
818 case -EBUSY:
819 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500820 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000821 case 0:
David Howellsea5b7782011-01-14 19:10:03 +0000822 dput(path->dentry);
823 if (*need_mntput)
824 mntput(path->mnt);
825 path->mnt = mnt;
826 path->dentry = dget(mnt->mnt_root);
827 *need_mntput = true;
828 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500829 default:
830 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000831 }
Al Viro19a167a2011-01-17 01:35:23 -0500832
David Howells9875cf82011-01-14 18:45:21 +0000833}
834
835/*
836 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000837 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000838 * - Flagged as mountpoint
839 * - Flagged as automount point
840 *
841 * This may only be called in refwalk mode.
842 *
843 * Serialization is taken care of in namespace.c
844 */
845static int follow_managed(struct path *path, unsigned flags)
846{
847 unsigned managed;
848 bool need_mntput = false;
849 int ret;
850
851 /* Given that we're not holding a lock here, we retain the value in a
852 * local variable for each dentry as we look at it so that we don't see
853 * the components of that value change under us */
854 while (managed = ACCESS_ONCE(path->dentry->d_flags),
855 managed &= DCACHE_MANAGED_DENTRY,
856 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000857 /* Allow the filesystem to manage the transit without i_mutex
858 * being held. */
859 if (managed & DCACHE_MANAGE_TRANSIT) {
860 BUG_ON(!path->dentry->d_op);
861 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400862 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000863 if (ret < 0)
864 return ret == -EISDIR ? 0 : ret;
865 }
866
David Howells9875cf82011-01-14 18:45:21 +0000867 /* Transit to a mounted filesystem. */
868 if (managed & DCACHE_MOUNTED) {
869 struct vfsmount *mounted = lookup_mnt(path);
870 if (mounted) {
871 dput(path->dentry);
872 if (need_mntput)
873 mntput(path->mnt);
874 path->mnt = mounted;
875 path->dentry = dget(mounted->mnt_root);
876 need_mntput = true;
877 continue;
878 }
879
880 /* Something is mounted on this dentry in another
881 * namespace and/or whatever was mounted there in this
882 * namespace got unmounted before we managed to get the
883 * vfsmount_lock */
884 }
885
886 /* Handle an automount point */
887 if (managed & DCACHE_NEED_AUTOMOUNT) {
888 ret = follow_automount(path, flags, &need_mntput);
889 if (ret < 0)
890 return ret == -EISDIR ? 0 : ret;
891 continue;
892 }
893
894 /* We didn't change the current path point */
895 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 }
David Howells9875cf82011-01-14 18:45:21 +0000897 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
David Howellscc53ce52011-01-14 18:45:26 +0000900int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 struct vfsmount *mounted;
903
Al Viro1c755af2009-04-18 14:06:57 -0400904 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400906 dput(path->dentry);
907 mntput(path->mnt);
908 path->mnt = mounted;
909 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 return 1;
911 }
912 return 0;
913}
914
Ian Kent62a73752011-03-25 01:51:02 +0800915static inline bool managed_dentry_might_block(struct dentry *dentry)
916{
917 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
918 dentry->d_op->d_manage(dentry, true) < 0);
919}
920
David Howells9875cf82011-01-14 18:45:21 +0000921/*
Al Viro287548e2011-05-27 06:50:06 -0400922 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
923 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000924 */
925static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400926 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000927{
Ian Kent62a73752011-03-25 01:51:02 +0800928 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +0000929 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800930 /*
931 * Don't forget we might have a non-mountpoint managed dentry
932 * that wants to block transit.
933 */
934 *inode = path->dentry->d_inode;
Al Viro287548e2011-05-27 06:50:06 -0400935 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000936 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800937
938 if (!d_mountpoint(path->dentry))
939 break;
940
David Howells9875cf82011-01-14 18:45:21 +0000941 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
942 if (!mounted)
943 break;
944 path->mnt = mounted;
945 path->dentry = mounted->mnt_root;
946 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
David Howells9875cf82011-01-14 18:45:21 +0000947 }
David Howells9875cf82011-01-14 18:45:21 +0000948 return true;
949}
950
Al Virodea39372011-05-27 06:53:39 -0400951static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400952{
Al Virodea39372011-05-27 06:53:39 -0400953 while (d_mountpoint(nd->path.dentry)) {
Al Viro287548e2011-05-27 06:50:06 -0400954 struct vfsmount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400955 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400956 if (!mounted)
957 break;
Al Virodea39372011-05-27 06:53:39 -0400958 nd->path.mnt = mounted;
959 nd->path.dentry = mounted->mnt_root;
960 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400961 }
962}
963
Nick Piggin31e6b012011-01-07 17:49:52 +1100964static int follow_dotdot_rcu(struct nameidata *nd)
965{
Nick Piggin31e6b012011-01-07 17:49:52 +1100966 set_root_rcu(nd);
967
David Howells9875cf82011-01-14 18:45:21 +0000968 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100969 if (nd->path.dentry == nd->root.dentry &&
970 nd->path.mnt == nd->root.mnt) {
971 break;
972 }
973 if (nd->path.dentry != nd->path.mnt->mnt_root) {
974 struct dentry *old = nd->path.dentry;
975 struct dentry *parent = old->d_parent;
976 unsigned seq;
977
978 seq = read_seqcount_begin(&parent->d_seq);
979 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500980 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100981 nd->path.dentry = parent;
982 nd->seq = seq;
983 break;
984 }
985 if (!follow_up_rcu(&nd->path))
986 break;
987 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100988 }
Al Virodea39372011-05-27 06:53:39 -0400989 follow_mount_rcu(nd);
990 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100991 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500992
993failed:
994 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500995 if (!(nd->flags & LOOKUP_ROOT))
996 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -0500997 rcu_read_unlock();
998 br_read_unlock(vfsmount_lock);
999 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001000}
1001
David Howells9875cf82011-01-14 18:45:21 +00001002/*
David Howellscc53ce52011-01-14 18:45:26 +00001003 * Follow down to the covering mount currently visible to userspace. At each
1004 * point, the filesystem owning that dentry may be queried as to whether the
1005 * caller is permitted to proceed or not.
1006 *
1007 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1008 * being true).
1009 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001010int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001011{
1012 unsigned managed;
1013 int ret;
1014
1015 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1016 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1017 /* Allow the filesystem to manage the transit without i_mutex
1018 * being held.
1019 *
1020 * We indicate to the filesystem if someone is trying to mount
1021 * something here. This gives autofs the chance to deny anyone
1022 * other than its daemon the right to mount on its
1023 * superstructure.
1024 *
1025 * The filesystem may sleep at this point.
1026 */
1027 if (managed & DCACHE_MANAGE_TRANSIT) {
1028 BUG_ON(!path->dentry->d_op);
1029 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001030 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001031 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001032 if (ret < 0)
1033 return ret == -EISDIR ? 0 : ret;
1034 }
1035
1036 /* Transit to a mounted filesystem. */
1037 if (managed & DCACHE_MOUNTED) {
1038 struct vfsmount *mounted = lookup_mnt(path);
1039 if (!mounted)
1040 break;
1041 dput(path->dentry);
1042 mntput(path->mnt);
1043 path->mnt = mounted;
1044 path->dentry = dget(mounted->mnt_root);
1045 continue;
1046 }
1047
1048 /* Don't handle automount points here */
1049 break;
1050 }
1051 return 0;
1052}
1053
1054/*
David Howells9875cf82011-01-14 18:45:21 +00001055 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1056 */
1057static void follow_mount(struct path *path)
1058{
1059 while (d_mountpoint(path->dentry)) {
1060 struct vfsmount *mounted = lookup_mnt(path);
1061 if (!mounted)
1062 break;
1063 dput(path->dentry);
1064 mntput(path->mnt);
1065 path->mnt = mounted;
1066 path->dentry = dget(mounted->mnt_root);
1067 }
1068}
1069
Nick Piggin31e6b012011-01-07 17:49:52 +11001070static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Al Viro2a737872009-04-07 11:49:53 -04001072 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001073
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001075 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Al Viro2a737872009-04-07 11:49:53 -04001077 if (nd->path.dentry == nd->root.dentry &&
1078 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 break;
1080 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001081 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001082 /* rare case of legitimate dget_parent()... */
1083 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 dput(old);
1085 break;
1086 }
Al Viro3088dd72010-01-30 15:47:29 -05001087 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 }
Al Viro79ed0222009-04-18 13:59:41 -04001090 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001091 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092}
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001095 * Allocate a dentry with name and parent, and perform a parent
1096 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1097 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1098 * have verified that no child exists while under i_mutex.
1099 */
1100static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1101 struct qstr *name, struct nameidata *nd)
1102{
1103 struct inode *inode = parent->d_inode;
1104 struct dentry *dentry;
1105 struct dentry *old;
1106
1107 /* Don't create child dentry for a dead directory. */
1108 if (unlikely(IS_DEADDIR(inode)))
1109 return ERR_PTR(-ENOENT);
1110
1111 dentry = d_alloc(parent, name);
1112 if (unlikely(!dentry))
1113 return ERR_PTR(-ENOMEM);
1114
1115 old = inode->i_op->lookup(inode, dentry, nd);
1116 if (unlikely(old)) {
1117 dput(dentry);
1118 dentry = old;
1119 }
1120 return dentry;
1121}
1122
1123/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 * It's more convoluted than I'd like it to be, but... it's still fairly
1125 * small and for now I'd prefer to have fast path as straight as possible.
1126 * It _is_ time-critical.
1127 */
1128static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001129 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130{
Jan Blunck4ac91372008-02-14 19:34:32 -08001131 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001132 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001133 int need_reval = 1;
1134 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001135 int err;
1136
Al Viro3cac2602009-08-13 18:27:43 +04001137 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001138 * Rename seqlock is not required here because in the off chance
1139 * of a false negative due to a concurrent rename, we're going to
1140 * do the non-racy lookup, below.
1141 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001142 if (nd->flags & LOOKUP_RCU) {
1143 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001144 *inode = nd->inode;
1145 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001146 if (!dentry)
1147 goto unlazy;
1148
Nick Piggin31e6b012011-01-07 17:49:52 +11001149 /* Memory barrier in read_seqcount_begin of child is enough */
1150 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1151 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001152 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001153
Al Viro24643082011-02-15 01:26:22 -05001154 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001155 status = d_revalidate(dentry, nd);
1156 if (unlikely(status <= 0)) {
1157 if (status != -ECHILD)
1158 need_reval = 0;
1159 goto unlazy;
1160 }
Al Viro24643082011-02-15 01:26:22 -05001161 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001162 path->mnt = mnt;
1163 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001164 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1165 goto unlazy;
1166 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1167 goto unlazy;
1168 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001169unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001170 if (unlazy_walk(nd, dentry))
1171 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001172 } else {
1173 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001174 }
Al Viro5a18fff2011-03-11 04:44:53 -05001175
1176retry:
1177 if (unlikely(!dentry)) {
1178 struct inode *dir = parent->d_inode;
1179 BUG_ON(nd->inode != dir);
1180
1181 mutex_lock(&dir->i_mutex);
1182 dentry = d_lookup(parent, name);
1183 if (likely(!dentry)) {
1184 dentry = d_alloc_and_lookup(parent, name, nd);
1185 if (IS_ERR(dentry)) {
1186 mutex_unlock(&dir->i_mutex);
1187 return PTR_ERR(dentry);
1188 }
1189 /* known good */
1190 need_reval = 0;
1191 status = 1;
1192 }
1193 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001194 }
Al Viro5a18fff2011-03-11 04:44:53 -05001195 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1196 status = d_revalidate(dentry, nd);
1197 if (unlikely(status <= 0)) {
1198 if (status < 0) {
1199 dput(dentry);
1200 return status;
1201 }
1202 if (!d_invalidate(dentry)) {
1203 dput(dentry);
1204 dentry = NULL;
1205 need_reval = 1;
1206 goto retry;
1207 }
1208 }
1209
David Howells9875cf82011-01-14 18:45:21 +00001210 path->mnt = mnt;
1211 path->dentry = dentry;
1212 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001213 if (unlikely(err < 0)) {
1214 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001215 return err;
Ian Kent89312212011-01-18 12:06:10 +08001216 }
David Howells9875cf82011-01-14 18:45:21 +00001217 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
Al Viro52094c82011-02-21 21:34:47 -05001221static inline int may_lookup(struct nameidata *nd)
1222{
1223 if (nd->flags & LOOKUP_RCU) {
1224 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1225 if (err != -ECHILD)
1226 return err;
Al Viro19660af2011-03-25 10:32:48 -04001227 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001228 return -ECHILD;
1229 }
1230 return exec_permission(nd->inode, 0);
1231}
1232
Al Viro9856fa12011-03-04 14:22:06 -05001233static inline int handle_dots(struct nameidata *nd, int type)
1234{
1235 if (type == LAST_DOTDOT) {
1236 if (nd->flags & LOOKUP_RCU) {
1237 if (follow_dotdot_rcu(nd))
1238 return -ECHILD;
1239 } else
1240 follow_dotdot(nd);
1241 }
1242 return 0;
1243}
1244
Al Viro951361f2011-03-04 14:44:37 -05001245static void terminate_walk(struct nameidata *nd)
1246{
1247 if (!(nd->flags & LOOKUP_RCU)) {
1248 path_put(&nd->path);
1249 } else {
1250 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001251 if (!(nd->flags & LOOKUP_ROOT))
1252 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001253 rcu_read_unlock();
1254 br_read_unlock(vfsmount_lock);
1255 }
1256}
1257
Al Viroce57dfc2011-03-13 19:58:58 -04001258static inline int walk_component(struct nameidata *nd, struct path *path,
1259 struct qstr *name, int type, int follow)
1260{
1261 struct inode *inode;
1262 int err;
1263 /*
1264 * "." and ".." are special - ".." especially so because it has
1265 * to be able to know about the current root directory and
1266 * parent relationships.
1267 */
1268 if (unlikely(type != LAST_NORM))
1269 return handle_dots(nd, type);
1270 err = do_lookup(nd, name, path, &inode);
1271 if (unlikely(err)) {
1272 terminate_walk(nd);
1273 return err;
1274 }
1275 if (!inode) {
1276 path_to_nameidata(path, nd);
1277 terminate_walk(nd);
1278 return -ENOENT;
1279 }
1280 if (unlikely(inode->i_op->follow_link) && follow) {
Al Viro19660af2011-03-25 10:32:48 -04001281 if (nd->flags & LOOKUP_RCU) {
1282 if (unlikely(unlazy_walk(nd, path->dentry))) {
1283 terminate_walk(nd);
1284 return -ECHILD;
1285 }
1286 }
Al Viroce57dfc2011-03-13 19:58:58 -04001287 BUG_ON(inode != path->dentry->d_inode);
1288 return 1;
1289 }
1290 path_to_nameidata(path, nd);
1291 nd->inode = inode;
1292 return 0;
1293}
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295/*
Al Virob3563792011-03-14 21:54:55 -04001296 * This limits recursive symlink follows to 8, while
1297 * limiting consecutive symlinks to 40.
1298 *
1299 * Without that kind of total limit, nasty chains of consecutive
1300 * symlinks can cause almost arbitrarily long lookups.
1301 */
1302static inline int nested_symlink(struct path *path, struct nameidata *nd)
1303{
1304 int res;
1305
Al Virob3563792011-03-14 21:54:55 -04001306 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1307 path_put_conditional(path, nd);
1308 path_put(&nd->path);
1309 return -ELOOP;
1310 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001311 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001312
1313 nd->depth++;
1314 current->link_count++;
1315
1316 do {
1317 struct path link = *path;
1318 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001319
1320 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001321 if (!res)
1322 res = walk_component(nd, path, &nd->last,
1323 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001324 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001325 } while (res > 0);
1326
1327 current->link_count--;
1328 nd->depth--;
1329 return res;
1330}
1331
1332/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001334 * This is the basic name resolution function, turning a pathname into
1335 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001337 * Returns 0 and nd will have valid dentry and mnt on success.
1338 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 */
Al Viro6de88d72009-08-09 01:41:57 +04001340static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 int err;
1344 unsigned int lookup_flags = nd->flags;
1345
1346 while (*name=='/')
1347 name++;
1348 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001349 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 /* At this point we know we have a real path component. */
1352 for(;;) {
1353 unsigned long hash;
1354 struct qstr this;
1355 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001356 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Trond Myklebustcdce5d62005-10-18 14:20:18 -07001358 nd->flags |= LOOKUP_CONTINUE;
Al Viro52094c82011-02-21 21:34:47 -05001359
1360 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 if (err)
1362 break;
1363
1364 this.name = name;
1365 c = *(const unsigned char *)name;
1366
1367 hash = init_name_hash();
1368 do {
1369 name++;
1370 hash = partial_name_hash(c, hash);
1371 c = *(const unsigned char *)name;
1372 } while (c && (c != '/'));
1373 this.len = name - (const char *) this.name;
1374 this.hash = end_name_hash(hash);
1375
Al Virofe479a52011-02-22 15:10:03 -05001376 type = LAST_NORM;
1377 if (this.name[0] == '.') switch (this.len) {
1378 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001379 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001380 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001381 nd->flags |= LOOKUP_JUMPED;
1382 }
Al Virofe479a52011-02-22 15:10:03 -05001383 break;
1384 case 1:
1385 type = LAST_DOT;
1386 }
Al Viro5a202bc2011-03-08 14:17:44 -05001387 if (likely(type == LAST_NORM)) {
1388 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001389 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001390 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1391 err = parent->d_op->d_hash(parent, nd->inode,
1392 &this);
1393 if (err < 0)
1394 break;
1395 }
1396 }
Al Virofe479a52011-02-22 15:10:03 -05001397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 /* remove trailing slashes? */
1399 if (!c)
1400 goto last_component;
1401 while (*++name == '/');
1402 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001403 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Al Viroce57dfc2011-03-13 19:58:58 -04001405 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1406 if (err < 0)
1407 return err;
Al Virofe479a52011-02-22 15:10:03 -05001408
Al Viroce57dfc2011-03-13 19:58:58 -04001409 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001410 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001412 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001415 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 break;
1417 continue;
1418 /* here ends the main loop */
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -08001421 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1422 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Al Virob3563792011-03-14 21:54:55 -04001423 nd->last = this;
1424 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001425 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 }
Al Viro951361f2011-03-04 14:44:37 -05001427 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 return err;
1429}
1430
Al Viro70e9b352011-03-05 21:12:22 -05001431static int path_init(int dfd, const char *name, unsigned int flags,
1432 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001434 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001435 int fput_needed;
1436 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001439 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001441 if (flags & LOOKUP_ROOT) {
1442 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001443 if (*name) {
1444 if (!inode->i_op->lookup)
1445 return -ENOTDIR;
1446 retval = inode_permission(inode, MAY_EXEC);
1447 if (retval)
1448 return retval;
1449 }
Al Viro5b6ca022011-03-09 23:04:47 -05001450 nd->path = nd->root;
1451 nd->inode = inode;
1452 if (flags & LOOKUP_RCU) {
1453 br_read_lock(vfsmount_lock);
1454 rcu_read_lock();
1455 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1456 } else {
1457 path_get(&nd->path);
1458 }
1459 return 0;
1460 }
1461
Al Viro2a737872009-04-07 11:49:53 -04001462 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001465 if (flags & LOOKUP_RCU) {
1466 br_read_lock(vfsmount_lock);
1467 rcu_read_lock();
1468 set_root_rcu(nd);
1469 } else {
1470 set_root(nd);
1471 path_get(&nd->root);
1472 }
Al Viro2a737872009-04-07 11:49:53 -04001473 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001474 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001475 if (flags & LOOKUP_RCU) {
1476 struct fs_struct *fs = current->fs;
1477 unsigned seq;
1478
1479 br_read_lock(vfsmount_lock);
1480 rcu_read_lock();
1481
1482 do {
1483 seq = read_seqcount_begin(&fs->seq);
1484 nd->path = fs->pwd;
1485 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1486 } while (read_seqcount_retry(&fs->seq, seq));
1487 } else {
1488 get_fs_pwd(current->fs, &nd->path);
1489 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001490 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001491 struct dentry *dentry;
1492
Al Viro1abf0c72011-03-13 03:51:11 -04001493 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001494 retval = -EBADF;
1495 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001496 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001497
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001498 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001499
Al Virof52e0c12011-03-14 18:56:51 -04001500 if (*name) {
1501 retval = -ENOTDIR;
1502 if (!S_ISDIR(dentry->d_inode->i_mode))
1503 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001504
Al Virof52e0c12011-03-14 18:56:51 -04001505 retval = file_permission(file, MAY_EXEC);
1506 if (retval)
1507 goto fput_fail;
1508 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001509
Jan Blunck5dd784d2008-02-14 19:34:38 -08001510 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001511 if (flags & LOOKUP_RCU) {
1512 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001513 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001514 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1515 br_read_lock(vfsmount_lock);
1516 rcu_read_lock();
1517 } else {
1518 path_get(&file->f_path);
1519 fput_light(file, fput_needed);
1520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 }
Al Viroe41f7d42011-02-22 14:02:58 -05001522
Nick Piggin31e6b012011-01-07 17:49:52 +11001523 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001524 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001525
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001526fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001527 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001528out_fail:
1529 return retval;
1530}
1531
Al Virobd92d7f2011-03-14 19:54:59 -04001532static inline int lookup_last(struct nameidata *nd, struct path *path)
1533{
1534 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1535 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1536
1537 nd->flags &= ~LOOKUP_PARENT;
1538 return walk_component(nd, path, &nd->last, nd->last_type,
1539 nd->flags & LOOKUP_FOLLOW);
1540}
1541
Al Viro9b4a9b12009-04-07 11:44:16 -04001542/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001543static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001544 unsigned int flags, struct nameidata *nd)
1545{
Al Viro70e9b352011-03-05 21:12:22 -05001546 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001547 struct path path;
1548 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001549
1550 /*
1551 * Path walking is largely split up into 2 different synchronisation
1552 * schemes, rcu-walk and ref-walk (explained in
1553 * Documentation/filesystems/path-lookup.txt). These share much of the
1554 * path walk code, but some things particularly setup, cleanup, and
1555 * following mounts are sufficiently divergent that functions are
1556 * duplicated. Typically there is a function foo(), and its RCU
1557 * analogue, foo_rcu().
1558 *
1559 * -ECHILD is the error number of choice (just to avoid clashes) that
1560 * is returned if some aspect of an rcu-walk fails. Such an error must
1561 * be handled by restarting a traditional ref-walk (which will always
1562 * be able to complete).
1563 */
Al Virobd92d7f2011-03-14 19:54:59 -04001564 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001565
Al Virobd92d7f2011-03-14 19:54:59 -04001566 if (unlikely(err))
1567 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001568
1569 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001570 err = link_path_walk(name, nd);
1571
1572 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001573 err = lookup_last(nd, &path);
1574 while (err > 0) {
1575 void *cookie;
1576 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001577 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001578 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001579 if (!err)
1580 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001581 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001582 }
1583 }
Al Viroee0827c2011-02-21 23:38:09 -05001584
Al Viro9f1fafe2011-03-25 11:00:12 -04001585 if (!err)
1586 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001587
1588 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1589 if (!nd->inode->i_op->lookup) {
1590 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001591 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001592 }
1593 }
Al Viro16c2cd72011-02-22 15:50:10 -05001594
Al Viro70e9b352011-03-05 21:12:22 -05001595 if (base)
1596 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001597
Al Viro5b6ca022011-03-09 23:04:47 -05001598 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001599 path_put(&nd->root);
1600 nd->root.mnt = NULL;
1601 }
Al Virobd92d7f2011-03-14 19:54:59 -04001602 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001603}
Nick Piggin31e6b012011-01-07 17:49:52 +11001604
Al Viroee0827c2011-02-21 23:38:09 -05001605static int do_path_lookup(int dfd, const char *name,
1606 unsigned int flags, struct nameidata *nd)
1607{
1608 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1609 if (unlikely(retval == -ECHILD))
1610 retval = path_lookupat(dfd, name, flags, nd);
1611 if (unlikely(retval == -ESTALE))
1612 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001613
1614 if (likely(!retval)) {
1615 if (unlikely(!audit_dummy_context())) {
1616 if (nd->path.dentry && nd->inode)
1617 audit_inode(name, nd->path.dentry);
1618 }
1619 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001620 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
1622
Al Viroc9c6cac2011-02-16 15:15:47 -05001623int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001624{
Al Viroc9c6cac2011-02-16 15:15:47 -05001625 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001626}
1627
Al Virod1811462008-08-02 00:49:18 -04001628int kern_path(const char *name, unsigned int flags, struct path *path)
1629{
1630 struct nameidata nd;
1631 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1632 if (!res)
1633 *path = nd.path;
1634 return res;
1635}
1636
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001637/**
1638 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1639 * @dentry: pointer to dentry of the base directory
1640 * @mnt: pointer to vfs mount of the base directory
1641 * @name: pointer to file name
1642 * @flags: lookup flags
1643 * @nd: pointer to nameidata
1644 */
1645int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1646 const char *name, unsigned int flags,
1647 struct nameidata *nd)
1648{
Al Viro5b6ca022011-03-09 23:04:47 -05001649 nd->root.dentry = dentry;
1650 nd->root.mnt = mnt;
1651 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1652 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001653}
1654
Christoph Hellwigeead1912007-10-16 23:25:38 -07001655static struct dentry *__lookup_hash(struct qstr *name,
1656 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001658 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001659 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 int err;
1661
Nick Pigginb74c79e2011-01-07 17:49:58 +11001662 err = exec_permission(inode, 0);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001663 if (err)
1664 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001667 * Don't bother with __d_lookup: callers are for creat as
1668 * well as unlink, so a lot of the time it would cost
1669 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001670 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001671 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001672
Nick Pigginfb045ad2011-01-07 17:49:55 +11001673 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro6e6b1bd2009-08-13 23:38:37 +04001674 dentry = do_revalidate(dentry, nd);
1675
Nick Pigginbaa03892010-08-18 04:37:31 +10001676 if (!dentry)
1677 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001678
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 return dentry;
1680}
1681
James Morris057f6c02007-04-26 00:12:05 -07001682/*
1683 * Restricted form of lookup. Doesn't follow links, single-component only,
1684 * needs parent already locked. Doesn't follow mounts.
1685 * SMP-safe.
1686 */
Adrian Bunka244e162006-03-31 02:32:11 -08001687static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688{
Jan Blunck4ac91372008-02-14 19:34:32 -08001689 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690}
1691
Christoph Hellwigeead1912007-10-16 23:25:38 -07001692/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001693 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001694 * @name: pathname component to lookup
1695 * @base: base directory to lookup from
1696 * @len: maximum length @len should be interpreted to
1697 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001698 * Note that this routine is purely a helper for filesystem usage and should
1699 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001700 * nameidata argument is passed to the filesystem methods and a filesystem
1701 * using this helper needs to be prepared for that.
1702 */
James Morris057f6c02007-04-26 00:12:05 -07001703struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1704{
James Morris057f6c02007-04-26 00:12:05 -07001705 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001706 unsigned long hash;
1707 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001708
David Woodhouse2f9092e2009-04-20 23:18:37 +01001709 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1710
Al Viro6a96ba52011-03-07 23:49:20 -05001711 this.name = name;
1712 this.len = len;
1713 if (!len)
1714 return ERR_PTR(-EACCES);
1715
1716 hash = init_name_hash();
1717 while (len--) {
1718 c = *(const unsigned char *)name++;
1719 if (c == '/' || c == '\0')
1720 return ERR_PTR(-EACCES);
1721 hash = partial_name_hash(c, hash);
1722 }
1723 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001724 /*
1725 * See if the low-level filesystem might want
1726 * to use its own hash..
1727 */
1728 if (base->d_flags & DCACHE_OP_HASH) {
1729 int err = base->d_op->d_hash(base, base->d_inode, &this);
1730 if (err < 0)
1731 return ERR_PTR(err);
1732 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001733
Christoph Hellwig49705b72005-11-08 21:35:06 -08001734 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001735}
1736
Al Viro2d8f3032008-07-22 09:59:21 -04001737int user_path_at(int dfd, const char __user *name, unsigned flags,
1738 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
Al Viro2d8f3032008-07-22 09:59:21 -04001740 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001741 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001744
1745 BUG_ON(flags & LOOKUP_PARENT);
1746
1747 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001749 if (!err)
1750 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
1752 return err;
1753}
1754
Al Viro2ad94ae2008-07-21 09:32:51 -04001755static int user_path_parent(int dfd, const char __user *path,
1756 struct nameidata *nd, char **name)
1757{
1758 char *s = getname(path);
1759 int error;
1760
1761 if (IS_ERR(s))
1762 return PTR_ERR(s);
1763
1764 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1765 if (error)
1766 putname(s);
1767 else
1768 *name = s;
1769
1770 return error;
1771}
1772
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773/*
1774 * It's inline, so penalty for filesystems that don't use sticky bit is
1775 * minimal.
1776 */
1777static inline int check_sticky(struct inode *dir, struct inode *inode)
1778{
David Howellsda9592e2008-11-14 10:39:05 +11001779 uid_t fsuid = current_fsuid();
1780
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 if (!(dir->i_mode & S_ISVTX))
1782 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001783 if (current_user_ns() != inode_userns(inode))
1784 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001785 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001787 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001789
1790other_userns:
1791 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792}
1793
1794/*
1795 * Check whether we can remove a link victim from directory dir, check
1796 * whether the type of victim is right.
1797 * 1. We can't do it if dir is read-only (done in permission())
1798 * 2. We should have write and exec permissions on dir
1799 * 3. We can't remove anything from append-only dir
1800 * 4. We can't do anything with immutable dir (done in permission())
1801 * 5. If the sticky bit on dir is set we should either
1802 * a. be owner of dir, or
1803 * b. be owner of victim, or
1804 * c. have CAP_FOWNER capability
1805 * 6. If the victim is append-only or immutable we can't do antyhing with
1806 * links pointing to it.
1807 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1808 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1809 * 9. We can't remove a root or mountpoint.
1810 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1811 * nfs_async_unlink().
1812 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001813static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814{
1815 int error;
1816
1817 if (!victim->d_inode)
1818 return -ENOENT;
1819
1820 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001821 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Al Virof419a2e2008-07-22 00:07:17 -04001823 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 if (error)
1825 return error;
1826 if (IS_APPEND(dir))
1827 return -EPERM;
1828 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001829 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 return -EPERM;
1831 if (isdir) {
1832 if (!S_ISDIR(victim->d_inode->i_mode))
1833 return -ENOTDIR;
1834 if (IS_ROOT(victim))
1835 return -EBUSY;
1836 } else if (S_ISDIR(victim->d_inode->i_mode))
1837 return -EISDIR;
1838 if (IS_DEADDIR(dir))
1839 return -ENOENT;
1840 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1841 return -EBUSY;
1842 return 0;
1843}
1844
1845/* Check whether we can create an object with dentry child in directory
1846 * dir.
1847 * 1. We can't do it if child already exists (open has special treatment for
1848 * this case, but since we are inlined it's OK)
1849 * 2. We can't do it if dir is read-only (done in permission())
1850 * 3. We should have write and exec permissions on dir
1851 * 4. We can't do it if dir is immutable (done in permission())
1852 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001853static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854{
1855 if (child->d_inode)
1856 return -EEXIST;
1857 if (IS_DEADDIR(dir))
1858 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001859 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860}
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862/*
1863 * p1 and p2 should be directories on the same fs.
1864 */
1865struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1866{
1867 struct dentry *p;
1868
1869 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001870 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 return NULL;
1872 }
1873
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001874 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001876 p = d_ancestor(p2, p1);
1877 if (p) {
1878 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1879 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1880 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
1882
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001883 p = d_ancestor(p1, p2);
1884 if (p) {
1885 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1886 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1887 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 }
1889
Ingo Molnarf2eace22006-07-03 00:25:05 -07001890 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1891 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 return NULL;
1893}
1894
1895void unlock_rename(struct dentry *p1, struct dentry *p2)
1896{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001897 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001899 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001900 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 }
1902}
1903
1904int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1905 struct nameidata *nd)
1906{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001907 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
1909 if (error)
1910 return error;
1911
Al Viroacfa4382008-12-04 10:06:33 -05001912 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 return -EACCES; /* shouldn't it be ENOSYS? */
1914 mode &= S_IALLUGO;
1915 mode |= S_IFREG;
1916 error = security_inode_create(dir, dentry, mode);
1917 if (error)
1918 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001920 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001921 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 return error;
1923}
1924
Al Viro73d049a2011-03-11 12:08:24 -05001925static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001927 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 struct inode *inode = dentry->d_inode;
1929 int error;
1930
Al Virobcda7652011-03-13 16:42:14 -04001931 /* O_PATH? */
1932 if (!acc_mode)
1933 return 0;
1934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 if (!inode)
1936 return -ENOENT;
1937
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001938 switch (inode->i_mode & S_IFMT) {
1939 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001941 case S_IFDIR:
1942 if (acc_mode & MAY_WRITE)
1943 return -EISDIR;
1944 break;
1945 case S_IFBLK:
1946 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001947 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001949 /*FALLTHRU*/
1950 case S_IFIFO:
1951 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001953 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001954 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001955
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001956 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001957 if (error)
1958 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /*
1961 * An append-only file must be opened in append mode for writing.
1962 */
1963 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001964 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001965 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001967 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 }
1969
1970 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001971 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05001972 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974 /*
1975 * Ensure there are no outstanding leases on the file.
1976 */
Al Virob65a9cf2009-12-16 06:27:40 -05001977 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001978}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Jeff Laytone1181ee2010-12-07 16:19:50 -05001980static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05001981{
Jeff Laytone1181ee2010-12-07 16:19:50 -05001982 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05001983 struct inode *inode = path->dentry->d_inode;
1984 int error = get_write_access(inode);
1985 if (error)
1986 return error;
1987 /*
1988 * Refuse to truncate files with mandatory locks held on them.
1989 */
1990 error = locks_verify_locked(inode);
1991 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09001992 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05001993 if (!error) {
1994 error = do_truncate(path->dentry, 0,
1995 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05001996 filp);
Al Viro7715b522009-12-16 03:54:00 -05001997 }
1998 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001999 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000}
2001
Dave Hansend57999e2008-02-15 14:37:27 -08002002/*
Dave Hansend57999e2008-02-15 14:37:27 -08002003 * Note that while the flag value (low two bits) for sys_open means:
2004 * 00 - read-only
2005 * 01 - write-only
2006 * 10 - read-write
2007 * 11 - special
2008 * it is changed into
2009 * 00 - no permissions needed
2010 * 01 - read-permission
2011 * 10 - write-permission
2012 * 11 - read-write
2013 * for the internal routines (ie open_namei()/follow_link() etc)
2014 * This is more logical, and also allows the 00 "no perm needed"
2015 * to be used for symlinks (where the permissions are checked
2016 * later).
2017 *
2018*/
2019static inline int open_to_namei_flags(int flag)
2020{
2021 if ((flag+1) & O_ACCMODE)
2022 flag++;
2023 return flag;
2024}
2025
Nick Piggin31e6b012011-01-07 17:49:52 +11002026/*
Al Virofe2d35f2011-03-05 22:58:25 -05002027 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002028 */
Al Virofb1cc552009-12-24 01:58:28 -05002029static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002030 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002031{
Al Viroa1e28032009-12-24 02:12:06 -05002032 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002033 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002034 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002035 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002036 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002037 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002038 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002039 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002040
Al Viroc3e380b2011-02-23 13:39:45 -05002041 nd->flags &= ~LOOKUP_PARENT;
2042 nd->flags |= op->intent;
2043
Al Viro1f36f772009-12-26 10:56:19 -05002044 switch (nd->last_type) {
2045 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002046 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002047 error = handle_dots(nd, nd->last_type);
2048 if (error)
2049 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002050 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002051 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002052 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002053 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002054 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002055 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002056 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002057 error = -EISDIR;
2058 goto exit;
2059 }
2060 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002061 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002062 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002063 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002064 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002065 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002066 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002067 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002068
Al Viroca344a892011-03-09 00:36:45 -05002069 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002070 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002071 if (nd->last.name[nd->last.len])
2072 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002073 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2074 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002075 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002076 error = walk_component(nd, path, &nd->last, LAST_NORM,
2077 !symlink_ok);
2078 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002079 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002080 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002081 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002082 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002083 error = complete_walk(nd);
2084 if (error)
2085 return ERR_PTR(-ECHILD);
Al Virofe2d35f2011-03-05 22:58:25 -05002086
2087 error = -ENOTDIR;
2088 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002089 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002090 goto exit;
2091 }
2092 audit_inode(pathname, nd->path.dentry);
2093 goto ok;
2094 }
2095
2096 /* create side of things */
Al Viro9f1fafe2011-03-25 11:00:12 -04002097 error = complete_walk(nd);
2098 if (error)
2099 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002100
2101 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002102 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002103 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002104 if (nd->last.name[nd->last.len])
2105 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002106
Al Viroa1e28032009-12-24 02:12:06 -05002107 mutex_lock(&dir->d_inode->i_mutex);
2108
Al Viro6c0d46c2011-03-09 00:59:59 -05002109 dentry = lookup_hash(nd);
2110 error = PTR_ERR(dentry);
2111 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002112 mutex_unlock(&dir->d_inode->i_mutex);
2113 goto exit;
2114 }
2115
Al Viro6c0d46c2011-03-09 00:59:59 -05002116 path->dentry = dentry;
2117 path->mnt = nd->path.mnt;
2118
Al Virofb1cc552009-12-24 01:58:28 -05002119 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002120 if (!dentry->d_inode) {
2121 int mode = op->mode;
2122 if (!IS_POSIXACL(dir->d_inode))
2123 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002124 /*
2125 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002126 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002127 * the time when the file is created and when
2128 * a permanent write count is taken through
2129 * the 'struct file' in nameidata_to_filp().
2130 */
2131 error = mnt_want_write(nd->path.mnt);
2132 if (error)
2133 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002134 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002135 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002136 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002137 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002138 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002139 error = security_path_mknod(&nd->path, dentry, mode, 0);
2140 if (error)
2141 goto exit_mutex_unlock;
2142 error = vfs_create(dir->d_inode, dentry, mode, nd);
2143 if (error)
2144 goto exit_mutex_unlock;
2145 mutex_unlock(&dir->d_inode->i_mutex);
2146 dput(nd->path.dentry);
2147 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002148 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002149 }
2150
2151 /*
2152 * It already exists.
2153 */
2154 mutex_unlock(&dir->d_inode->i_mutex);
2155 audit_inode(pathname, path->dentry);
2156
2157 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002158 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002159 goto exit_dput;
2160
David Howells9875cf82011-01-14 18:45:21 +00002161 error = follow_managed(path, nd->flags);
2162 if (error < 0)
2163 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002164
2165 error = -ENOENT;
2166 if (!path->dentry->d_inode)
2167 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002168
2169 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002170 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002171
2172 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002173 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002174 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002175 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002176 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002177ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002178 if (!S_ISREG(nd->inode->i_mode))
2179 will_truncate = 0;
2180
Al Viro0f9d1a12011-03-09 00:13:14 -05002181 if (will_truncate) {
2182 error = mnt_want_write(nd->path.mnt);
2183 if (error)
2184 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002185 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002186 }
Al Viroca344a892011-03-09 00:36:45 -05002187common:
Al Virobcda7652011-03-13 16:42:14 -04002188 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002189 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002190 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002191 filp = nameidata_to_filp(nd);
2192 if (!IS_ERR(filp)) {
2193 error = ima_file_check(filp, op->acc_mode);
2194 if (error) {
2195 fput(filp);
2196 filp = ERR_PTR(error);
2197 }
2198 }
2199 if (!IS_ERR(filp)) {
2200 if (will_truncate) {
2201 error = handle_truncate(filp);
2202 if (error) {
2203 fput(filp);
2204 filp = ERR_PTR(error);
2205 }
2206 }
2207 }
Al Viroca344a892011-03-09 00:36:45 -05002208out:
2209 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002210 mnt_drop_write(nd->path.mnt);
2211 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002212 return filp;
2213
2214exit_mutex_unlock:
2215 mutex_unlock(&dir->d_inode->i_mutex);
2216exit_dput:
2217 path_put_conditional(path, nd);
2218exit:
Al Viroca344a892011-03-09 00:36:45 -05002219 filp = ERR_PTR(error);
2220 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002221}
2222
Al Viro13aab422011-02-23 17:54:08 -05002223static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002224 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225{
Al Virofe2d35f2011-03-05 22:58:25 -05002226 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002227 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002228 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002229 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002230
2231 filp = get_empty_filp();
2232 if (!filp)
2233 return ERR_PTR(-ENFILE);
2234
Al Viro47c805d2011-02-23 17:44:09 -05002235 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002236 nd->intent.open.file = filp;
2237 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2238 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002239
Al Viro73d049a2011-03-11 12:08:24 -05002240 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002241 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002242 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002243
Al Virofe2d35f2011-03-05 22:58:25 -05002244 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002245 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002246 if (unlikely(error))
2247 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
Al Viro73d049a2011-03-11 12:08:24 -05002249 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002250 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002251 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002252 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002253 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002254 path_put_conditional(&path, nd);
2255 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002256 filp = ERR_PTR(-ELOOP);
2257 break;
2258 }
Al Viro73d049a2011-03-11 12:08:24 -05002259 nd->flags |= LOOKUP_PARENT;
2260 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002261 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002262 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002263 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002264 else
Al Viro73d049a2011-03-11 12:08:24 -05002265 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002266 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002267 }
Al Viro10fa8e62009-12-26 07:09:49 -05002268out:
Al Viro73d049a2011-03-11 12:08:24 -05002269 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2270 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002271 if (base)
2272 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002273 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002274 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
Nick Piggin31e6b012011-01-07 17:49:52 +11002276out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002277 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002278 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279}
2280
Al Viro13aab422011-02-23 17:54:08 -05002281struct file *do_filp_open(int dfd, const char *pathname,
2282 const struct open_flags *op, int flags)
2283{
Al Viro73d049a2011-03-11 12:08:24 -05002284 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002285 struct file *filp;
2286
Al Viro73d049a2011-03-11 12:08:24 -05002287 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002288 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002289 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002290 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002291 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002292 return filp;
2293}
2294
Al Viro73d049a2011-03-11 12:08:24 -05002295struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2296 const char *name, const struct open_flags *op, int flags)
2297{
2298 struct nameidata nd;
2299 struct file *file;
2300
2301 nd.root.mnt = mnt;
2302 nd.root.dentry = dentry;
2303
2304 flags |= LOOKUP_ROOT;
2305
Al Virobcda7652011-03-13 16:42:14 -04002306 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002307 return ERR_PTR(-ELOOP);
2308
2309 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2310 if (unlikely(file == ERR_PTR(-ECHILD)))
2311 file = path_openat(-1, name, &nd, op, flags);
2312 if (unlikely(file == ERR_PTR(-ESTALE)))
2313 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2314 return file;
2315}
2316
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317/**
2318 * lookup_create - lookup a dentry, creating it if it doesn't exist
2319 * @nd: nameidata info
2320 * @is_dir: directory flag
2321 *
2322 * Simple function to lookup and return a dentry and create it
2323 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002324 *
Jan Blunck4ac91372008-02-14 19:34:32 -08002325 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 */
2327struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2328{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002329 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
Jan Blunck4ac91372008-02-14 19:34:32 -08002331 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002332 /*
2333 * Yucky last component or no last component at all?
2334 * (foo/., foo/.., /////)
2335 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 if (nd->last_type != LAST_NORM)
2337 goto fail;
2338 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04002339 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04002340 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002341
2342 /*
2343 * Do the final lookup.
2344 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08002345 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 if (IS_ERR(dentry))
2347 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002348
Al Viroe9baf6e2008-05-15 04:49:12 -04002349 if (dentry->d_inode)
2350 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002351 /*
2352 * Special case - lookup gave negative, but... we had foo/bar/
2353 * From the vfs_mknod() POV we just have a negative dentry -
2354 * all is fine. Let's be bastards - you had / on the end, you've
2355 * been asking for (non-existent) directory. -ENOENT for you.
2356 */
Al Viroe9baf6e2008-05-15 04:49:12 -04002357 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2358 dput(dentry);
2359 dentry = ERR_PTR(-ENOENT);
2360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002362eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002364 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365fail:
2366 return dentry;
2367}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07002368EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
2370int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2371{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002372 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
2374 if (error)
2375 return error;
2376
Serge E. Hallyne795b712011-03-23 16:43:25 -07002377 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2378 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 return -EPERM;
2380
Al Viroacfa4382008-12-04 10:06:33 -05002381 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 return -EPERM;
2383
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002384 error = devcgroup_inode_mknod(mode, dev);
2385 if (error)
2386 return error;
2387
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 error = security_inode_mknod(dir, dentry, mode, dev);
2389 if (error)
2390 return error;
2391
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002393 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002394 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return error;
2396}
2397
Dave Hansen463c3192008-02-15 14:37:57 -08002398static int may_mknod(mode_t mode)
2399{
2400 switch (mode & S_IFMT) {
2401 case S_IFREG:
2402 case S_IFCHR:
2403 case S_IFBLK:
2404 case S_IFIFO:
2405 case S_IFSOCK:
2406 case 0: /* zero mode translates to S_IFREG */
2407 return 0;
2408 case S_IFDIR:
2409 return -EPERM;
2410 default:
2411 return -EINVAL;
2412 }
2413}
2414
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002415SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2416 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417{
Al Viro2ad94ae2008-07-21 09:32:51 -04002418 int error;
2419 char *tmp;
2420 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 struct nameidata nd;
2422
2423 if (S_ISDIR(mode))
2424 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Al Viro2ad94ae2008-07-21 09:32:51 -04002426 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002428 return error;
2429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002431 if (IS_ERR(dentry)) {
2432 error = PTR_ERR(dentry);
2433 goto out_unlock;
2434 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002435 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002436 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002437 error = may_mknod(mode);
2438 if (error)
2439 goto out_dput;
2440 error = mnt_want_write(nd.path.mnt);
2441 if (error)
2442 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002443 error = security_path_mknod(&nd.path, dentry, mode, dev);
2444 if (error)
2445 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002446 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002448 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 break;
2450 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002451 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 new_decode_dev(dev));
2453 break;
2454 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002455 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002458out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002459 mnt_drop_write(nd.path.mnt);
2460out_dput:
2461 dput(dentry);
2462out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002463 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002464 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 putname(tmp);
2466
2467 return error;
2468}
2469
Heiko Carstens3480b252009-01-14 14:14:16 +01002470SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002471{
2472 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2473}
2474
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2476{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002477 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
2479 if (error)
2480 return error;
2481
Al Viroacfa4382008-12-04 10:06:33 -05002482 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 return -EPERM;
2484
2485 mode &= (S_IRWXUGO|S_ISVTX);
2486 error = security_inode_mkdir(dir, dentry, mode);
2487 if (error)
2488 return error;
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002491 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002492 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 return error;
2494}
2495
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002496SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497{
2498 int error = 0;
2499 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002500 struct dentry *dentry;
2501 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Al Viro2ad94ae2008-07-21 09:32:51 -04002503 error = user_path_parent(dfd, pathname, &nd, &tmp);
2504 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002505 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506
Dave Hansen6902d922006-09-30 23:29:01 -07002507 dentry = lookup_create(&nd, 1);
2508 error = PTR_ERR(dentry);
2509 if (IS_ERR(dentry))
2510 goto out_unlock;
2511
Jan Blunck4ac91372008-02-14 19:34:32 -08002512 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002513 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002514 error = mnt_want_write(nd.path.mnt);
2515 if (error)
2516 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002517 error = security_path_mkdir(&nd.path, dentry, mode);
2518 if (error)
2519 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002520 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002521out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002522 mnt_drop_write(nd.path.mnt);
2523out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002524 dput(dentry);
2525out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002526 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002527 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002528 putname(tmp);
2529out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 return error;
2531}
2532
Heiko Carstens3cdad422009-01-14 14:14:22 +01002533SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002534{
2535 return sys_mkdirat(AT_FDCWD, pathname, mode);
2536}
2537
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538/*
Sage Weila71905f2011-05-24 13:06:08 -07002539 * The dentry_unhash() helper will try to drop the dentry early: we
2540 * should have a usage count of 2 if we're the only user of this
2541 * dentry, and if that is true (possibly after pruning the dcache),
2542 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 *
2544 * A low-level filesystem can, if it choses, legally
2545 * do a
2546 *
2547 * if (!d_unhashed(dentry))
2548 * return -EBUSY;
2549 *
2550 * if it cannot handle the case of removing a directory
2551 * that is still in use by something else..
2552 */
2553void dentry_unhash(struct dentry *dentry)
2554{
Vasily Averindc168422006-12-06 20:37:07 -08002555 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002557 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 __d_drop(dentry);
2559 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
2561
2562int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2563{
2564 int error = may_delete(dir, dentry, 1);
2565
2566 if (error)
2567 return error;
2568
Al Viroacfa4382008-12-04 10:06:33 -05002569 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 return -EPERM;
2571
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002572 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
Sage Weil912dbc12011-05-24 13:06:11 -07002574 error = -EBUSY;
2575 if (d_mountpoint(dentry))
2576 goto out;
2577
2578 error = security_inode_rmdir(dir, dentry);
2579 if (error)
2580 goto out;
2581
2582 error = dir->i_op->rmdir(dir, dentry);
2583 if (error)
2584 goto out;
2585
2586 dentry->d_inode->i_flags |= S_DEAD;
2587 dont_mount(dentry);
2588
2589out:
2590 mutex_unlock(&dentry->d_inode->i_mutex);
2591 if (!error)
2592 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 return error;
2594}
2595
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002596static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597{
2598 int error = 0;
2599 char * name;
2600 struct dentry *dentry;
2601 struct nameidata nd;
2602
Al Viro2ad94ae2008-07-21 09:32:51 -04002603 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002605 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606
2607 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002608 case LAST_DOTDOT:
2609 error = -ENOTEMPTY;
2610 goto exit1;
2611 case LAST_DOT:
2612 error = -EINVAL;
2613 goto exit1;
2614 case LAST_ROOT:
2615 error = -EBUSY;
2616 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002618
2619 nd.flags &= ~LOOKUP_PARENT;
2620
Jan Blunck4ac91372008-02-14 19:34:32 -08002621 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002622 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002624 if (IS_ERR(dentry))
2625 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002626 error = mnt_want_write(nd.path.mnt);
2627 if (error)
2628 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002629 error = security_path_rmdir(&nd.path, dentry);
2630 if (error)
2631 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002632 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002633exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002634 mnt_drop_write(nd.path.mnt);
2635exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002636 dput(dentry);
2637exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002638 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002640 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 putname(name);
2642 return error;
2643}
2644
Heiko Carstens3cdad422009-01-14 14:14:22 +01002645SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002646{
2647 return do_rmdir(AT_FDCWD, pathname);
2648}
2649
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650int vfs_unlink(struct inode *dir, struct dentry *dentry)
2651{
2652 int error = may_delete(dir, dentry, 0);
2653
2654 if (error)
2655 return error;
2656
Al Viroacfa4382008-12-04 10:06:33 -05002657 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 return -EPERM;
2659
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002660 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 if (d_mountpoint(dentry))
2662 error = -EBUSY;
2663 else {
2664 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002665 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002667 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002668 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002671 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
2673 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2674 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002675 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 }
Robert Love0eeca282005-07-12 17:06:03 -04002678
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 return error;
2680}
2681
2682/*
2683 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002684 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 * writeout happening, and we don't want to prevent access to the directory
2686 * while waiting on the I/O.
2687 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002688static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689{
Al Viro2ad94ae2008-07-21 09:32:51 -04002690 int error;
2691 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 struct dentry *dentry;
2693 struct nameidata nd;
2694 struct inode *inode = NULL;
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;
2699
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 error = -EISDIR;
2701 if (nd.last_type != LAST_NORM)
2702 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002703
2704 nd.flags &= ~LOOKUP_PARENT;
2705
Jan Blunck4ac91372008-02-14 19:34:32 -08002706 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002707 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 error = PTR_ERR(dentry);
2709 if (!IS_ERR(dentry)) {
2710 /* Why not before? Because we want correct error value */
2711 if (nd.last.name[nd.last.len])
2712 goto slashes;
2713 inode = dentry->d_inode;
2714 if (inode)
Al Viro7de9c6e2010-10-23 11:11:40 -04002715 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002716 error = mnt_want_write(nd.path.mnt);
2717 if (error)
2718 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002719 error = security_path_unlink(&nd.path, dentry);
2720 if (error)
2721 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002722 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002723exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002724 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 exit2:
2726 dput(dentry);
2727 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002728 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 if (inode)
2730 iput(inode); /* truncate the inode here */
2731exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002732 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 putname(name);
2734 return error;
2735
2736slashes:
2737 error = !dentry->d_inode ? -ENOENT :
2738 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2739 goto exit2;
2740}
2741
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002742SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002743{
2744 if ((flag & ~AT_REMOVEDIR) != 0)
2745 return -EINVAL;
2746
2747 if (flag & AT_REMOVEDIR)
2748 return do_rmdir(dfd, pathname);
2749
2750 return do_unlinkat(dfd, pathname);
2751}
2752
Heiko Carstens3480b252009-01-14 14:14:16 +01002753SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002754{
2755 return do_unlinkat(AT_FDCWD, pathname);
2756}
2757
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002758int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002760 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
2762 if (error)
2763 return error;
2764
Al Viroacfa4382008-12-04 10:06:33 -05002765 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 return -EPERM;
2767
2768 error = security_inode_symlink(dir, dentry, oldname);
2769 if (error)
2770 return error;
2771
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002773 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002774 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 return error;
2776}
2777
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002778SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2779 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780{
Al Viro2ad94ae2008-07-21 09:32:51 -04002781 int error;
2782 char *from;
2783 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002784 struct dentry *dentry;
2785 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786
2787 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002788 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002790
2791 error = user_path_parent(newdfd, newname, &nd, &to);
2792 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002793 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
Dave Hansen6902d922006-09-30 23:29:01 -07002795 dentry = lookup_create(&nd, 0);
2796 error = PTR_ERR(dentry);
2797 if (IS_ERR(dentry))
2798 goto out_unlock;
2799
Dave Hansen75c3f292008-02-15 14:37:45 -08002800 error = mnt_want_write(nd.path.mnt);
2801 if (error)
2802 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002803 error = security_path_symlink(&nd.path, dentry, from);
2804 if (error)
2805 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002806 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002807out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002808 mnt_drop_write(nd.path.mnt);
2809out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002810 dput(dentry);
2811out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002812 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002813 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002814 putname(to);
2815out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 putname(from);
2817 return error;
2818}
2819
Heiko Carstens3480b252009-01-14 14:14:16 +01002820SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002821{
2822 return sys_symlinkat(oldname, AT_FDCWD, newname);
2823}
2824
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2826{
2827 struct inode *inode = old_dentry->d_inode;
2828 int error;
2829
2830 if (!inode)
2831 return -ENOENT;
2832
Miklos Szeredia95164d2008-07-30 15:08:48 +02002833 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 if (error)
2835 return error;
2836
2837 if (dir->i_sb != inode->i_sb)
2838 return -EXDEV;
2839
2840 /*
2841 * A link to an append-only or immutable file cannot be created.
2842 */
2843 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2844 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002845 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002847 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 return -EPERM;
2849
2850 error = security_inode_link(old_dentry, dir, new_dentry);
2851 if (error)
2852 return error;
2853
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002854 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302855 /* Make sure we don't allow creating hardlink to an unlinked file */
2856 if (inode->i_nlink == 0)
2857 error = -ENOENT;
2858 else
2859 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002860 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002861 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002862 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 return error;
2864}
2865
2866/*
2867 * Hardlinks are often used in delicate situations. We avoid
2868 * security-related surprises by not following symlinks on the
2869 * newname. --KAB
2870 *
2871 * We don't follow them on the oldname either to be compatible
2872 * with linux 2.0, and to avoid hard-linking to directories
2873 * and other special files. --ADM
2874 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002875SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2876 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877{
2878 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002879 struct nameidata nd;
2880 struct path old_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302881 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002883 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302885 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002886 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302887 /*
2888 * To use null names we require CAP_DAC_READ_SEARCH
2889 * This ensures that not everyone will be able to create
2890 * handlink using the passed filedescriptor.
2891 */
2892 if (flags & AT_EMPTY_PATH) {
2893 if (!capable(CAP_DAC_READ_SEARCH))
2894 return -ENOENT;
2895 how = LOOKUP_EMPTY;
2896 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002897
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302898 if (flags & AT_SYMLINK_FOLLOW)
2899 how |= LOOKUP_FOLLOW;
2900
2901 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002903 return error;
2904
2905 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 if (error)
2907 goto out;
2908 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002909 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 goto out_release;
2911 new_dentry = lookup_create(&nd, 0);
2912 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002913 if (IS_ERR(new_dentry))
2914 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002915 error = mnt_want_write(nd.path.mnt);
2916 if (error)
2917 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002918 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2919 if (error)
2920 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002921 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002922out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002923 mnt_drop_write(nd.path.mnt);
2924out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002925 dput(new_dentry);
2926out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002927 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002929 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002930 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931out:
Al Viro2d8f3032008-07-22 09:59:21 -04002932 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933
2934 return error;
2935}
2936
Heiko Carstens3480b252009-01-14 14:14:16 +01002937SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002938{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002939 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002940}
2941
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942/*
2943 * The worst of all namespace operations - renaming directory. "Perverted"
2944 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2945 * Problems:
2946 * a) we can get into loop creation. Check is done in is_subdir().
2947 * b) race potential - two innocent renames can create a loop together.
2948 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002949 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 * story.
2951 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002952 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 * whether the target exists). Solution: try to be smart with locking
2954 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002955 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 * move will be locked. Thus we can rank directories by the tree
2957 * (ancestors first) and rank all non-directories after them.
2958 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002959 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 * HOWEVER, it relies on the assumption that any object with ->lookup()
2961 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2962 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07002963 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002964 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002966 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 * locking].
2968 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002969static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2970 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971{
2972 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07002973 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974
2975 /*
2976 * If we are going to change the parent - check write permissions,
2977 * we'll need to flip '..'.
2978 */
2979 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002980 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 if (error)
2982 return error;
2983 }
2984
2985 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2986 if (error)
2987 return error;
2988
Al Virod83c49f2010-04-30 17:17:09 -04002989 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002990 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07002991
2992 error = -EBUSY;
2993 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2994 goto out;
2995
2996 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2997 if (error)
2998 goto out;
2999
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003001 target->i_flags |= S_DEAD;
3002 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 }
Sage Weil9055cba2011-05-24 13:06:12 -07003004out:
3005 if (target)
3006 mutex_unlock(&target->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003007 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003008 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3009 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 return error;
3011}
3012
Adrian Bunk75c96f82005-05-05 16:16:09 -07003013static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3014 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015{
Sage Weil51892bb2011-05-24 13:06:13 -07003016 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 int error;
3018
3019 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3020 if (error)
3021 return error;
3022
3023 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003025 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003026
3027 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003029 goto out;
3030
3031 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3032 if (error)
3033 goto out;
3034
3035 if (target)
3036 dont_mount(new_dentry);
3037 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3038 d_move(old_dentry, new_dentry);
3039out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003041 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 dput(new_dentry);
3043 return error;
3044}
3045
3046int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3047 struct inode *new_dir, struct dentry *new_dentry)
3048{
3049 int error;
3050 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003051 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
3053 if (old_dentry->d_inode == new_dentry->d_inode)
3054 return 0;
3055
3056 error = may_delete(old_dir, old_dentry, is_dir);
3057 if (error)
3058 return error;
3059
3060 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003061 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062 else
3063 error = may_delete(new_dir, new_dentry, is_dir);
3064 if (error)
3065 return error;
3066
Al Viroacfa4382008-12-04 10:06:33 -05003067 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 return -EPERM;
3069
Robert Love0eeca282005-07-12 17:06:03 -04003070 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3071
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 if (is_dir)
3073 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3074 else
3075 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003076 if (!error)
3077 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003078 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003079 fsnotify_oldname_free(old_name);
3080
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 return error;
3082}
3083
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003084SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3085 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086{
Al Viro2ad94ae2008-07-21 09:32:51 -04003087 struct dentry *old_dir, *new_dir;
3088 struct dentry *old_dentry, *new_dentry;
3089 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003091 char *from;
3092 char *to;
3093 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094
Al Viro2ad94ae2008-07-21 09:32:51 -04003095 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 if (error)
3097 goto exit;
3098
Al Viro2ad94ae2008-07-21 09:32:51 -04003099 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 if (error)
3101 goto exit1;
3102
3103 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003104 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 goto exit2;
3106
Jan Blunck4ac91372008-02-14 19:34:32 -08003107 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 error = -EBUSY;
3109 if (oldnd.last_type != LAST_NORM)
3110 goto exit2;
3111
Jan Blunck4ac91372008-02-14 19:34:32 -08003112 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 if (newnd.last_type != LAST_NORM)
3114 goto exit2;
3115
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003116 oldnd.flags &= ~LOOKUP_PARENT;
3117 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003118 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003119
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 trap = lock_rename(new_dir, old_dir);
3121
Christoph Hellwig49705b72005-11-08 21:35:06 -08003122 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 error = PTR_ERR(old_dentry);
3124 if (IS_ERR(old_dentry))
3125 goto exit3;
3126 /* source must exist */
3127 error = -ENOENT;
3128 if (!old_dentry->d_inode)
3129 goto exit4;
3130 /* unless the source is a directory trailing slashes give -ENOTDIR */
3131 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3132 error = -ENOTDIR;
3133 if (oldnd.last.name[oldnd.last.len])
3134 goto exit4;
3135 if (newnd.last.name[newnd.last.len])
3136 goto exit4;
3137 }
3138 /* source should not be ancestor of target */
3139 error = -EINVAL;
3140 if (old_dentry == trap)
3141 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003142 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 error = PTR_ERR(new_dentry);
3144 if (IS_ERR(new_dentry))
3145 goto exit4;
3146 /* target should not be an ancestor of source */
3147 error = -ENOTEMPTY;
3148 if (new_dentry == trap)
3149 goto exit5;
3150
Dave Hansen9079b1e2008-02-15 14:37:49 -08003151 error = mnt_want_write(oldnd.path.mnt);
3152 if (error)
3153 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003154 error = security_path_rename(&oldnd.path, old_dentry,
3155 &newnd.path, new_dentry);
3156 if (error)
3157 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 error = vfs_rename(old_dir->d_inode, old_dentry,
3159 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003160exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003161 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162exit5:
3163 dput(new_dentry);
3164exit4:
3165 dput(old_dentry);
3166exit3:
3167 unlock_rename(new_dir, old_dir);
3168exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003169 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003170 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003172 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003174exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 return error;
3176}
3177
Heiko Carstensa26eab22009-01-14 14:14:17 +01003178SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003179{
3180 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3181}
3182
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3184{
3185 int len;
3186
3187 len = PTR_ERR(link);
3188 if (IS_ERR(link))
3189 goto out;
3190
3191 len = strlen(link);
3192 if (len > (unsigned) buflen)
3193 len = buflen;
3194 if (copy_to_user(buffer, link, len))
3195 len = -EFAULT;
3196out:
3197 return len;
3198}
3199
3200/*
3201 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3202 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3203 * using) it for any given inode is up to filesystem.
3204 */
3205int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3206{
3207 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003208 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003209 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003210
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003212 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003213 if (IS_ERR(cookie))
3214 return PTR_ERR(cookie);
3215
3216 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3217 if (dentry->d_inode->i_op->put_link)
3218 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3219 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220}
3221
3222int vfs_follow_link(struct nameidata *nd, const char *link)
3223{
3224 return __vfs_follow_link(nd, link);
3225}
3226
3227/* get the link contents into pagecache */
3228static char *page_getlink(struct dentry * dentry, struct page **ppage)
3229{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003230 char *kaddr;
3231 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003233 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003235 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003237 kaddr = kmap(page);
3238 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3239 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240}
3241
3242int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3243{
3244 struct page *page = NULL;
3245 char *s = page_getlink(dentry, &page);
3246 int res = vfs_readlink(dentry,buffer,buflen,s);
3247 if (page) {
3248 kunmap(page);
3249 page_cache_release(page);
3250 }
3251 return res;
3252}
3253
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003254void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003256 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003258 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259}
3260
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003261void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003263 struct page *page = cookie;
3264
3265 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 kunmap(page);
3267 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 }
3269}
3270
Nick Piggin54566b22009-01-04 12:00:53 -08003271/*
3272 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3273 */
3274int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275{
3276 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003277 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003278 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003279 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003281 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3282 if (nofs)
3283 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284
NeilBrown7e53cac2006-03-25 03:07:57 -08003285retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003286 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003287 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003289 goto fail;
3290
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 kaddr = kmap_atomic(page, KM_USER0);
3292 memcpy(kaddr, symname, len-1);
3293 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003294
3295 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3296 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 if (err < 0)
3298 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003299 if (err < len-1)
3300 goto retry;
3301
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 mark_inode_dirty(inode);
3303 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304fail:
3305 return err;
3306}
3307
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003308int page_symlink(struct inode *inode, const char *symname, int len)
3309{
3310 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003311 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003312}
3313
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003314const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 .readlink = generic_readlink,
3316 .follow_link = page_follow_link_light,
3317 .put_link = page_put_link,
3318};
3319
Al Viro2d8f3032008-07-22 09:59:21 -04003320EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003321EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322EXPORT_SYMBOL(follow_down);
3323EXPORT_SYMBOL(follow_up);
3324EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3325EXPORT_SYMBOL(getname);
3326EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327EXPORT_SYMBOL(lookup_one_len);
3328EXPORT_SYMBOL(page_follow_link_light);
3329EXPORT_SYMBOL(page_put_link);
3330EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003331EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332EXPORT_SYMBOL(page_symlink);
3333EXPORT_SYMBOL(page_symlink_inode_operations);
Al Viroc9c6cac2011-02-16 15:15:47 -05003334EXPORT_SYMBOL(kern_path_parent);
Al Virod1811462008-08-02 00:49:18 -04003335EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003336EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003337EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08003338EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339EXPORT_SYMBOL(unlock_rename);
3340EXPORT_SYMBOL(vfs_create);
3341EXPORT_SYMBOL(vfs_follow_link);
3342EXPORT_SYMBOL(vfs_link);
3343EXPORT_SYMBOL(vfs_mkdir);
3344EXPORT_SYMBOL(vfs_mknod);
3345EXPORT_SYMBOL(generic_permission);
3346EXPORT_SYMBOL(vfs_readlink);
3347EXPORT_SYMBOL(vfs_rename);
3348EXPORT_SYMBOL(vfs_rmdir);
3349EXPORT_SYMBOL(vfs_symlink);
3350EXPORT_SYMBOL(vfs_unlink);
3351EXPORT_SYMBOL(dentry_unhash);
3352EXPORT_SYMBOL(generic_readlink);