blob: def63e7c058dcc50cd4c72fa400536b71e72ebce [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
Eric Parise81e3f42009-12-04 15:47:36 -050037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800117static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
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) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400161 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700169/*
170 * This does basic POSIX ACL permission checking
171 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100172static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700174{
175 umode_t mode = inode->i_mode;
176
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
179 if (current_fsuid() == inode->i_uid)
180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
185 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
195 if ((mask & ~mode) == 0)
196 return 0;
197 return -EACCES;
198}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100201 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
Randy Dunlap39191622011-01-08 19:36:21 -0800205 * @flags: IPERM_FLAG_ flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 *
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100210 * are used for other things.
211 *
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100216int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700219 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700222 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100224 ret = acl_permission_check(inode, mask, flags, check_acl);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700225 if (ret != -EACCES)
226 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 /*
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
231 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200232 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (capable(CAP_DAC_OVERRIDE))
234 return 0;
235
236 /*
237 * Searching includes executable on directories, else just read.
238 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
242 return 0;
243
244 return -EACCES;
245}
246
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200247/**
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
251 *
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
256 */
Al Virof419a2e2008-07-22 00:07:17 -0400257int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Al Viroe6305c42008-07-15 21:03:57 -0400259 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700262 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
264 /*
265 * Nobody gets write access to a read-only fs.
266 */
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
269 return -EROFS;
270
271 /*
272 * Nobody gets write access to an immutable file.
273 */
274 if (IS_IMMUTABLE(inode))
275 return -EACCES;
276 }
277
Al Viroacfa4382008-12-04 10:06:33 -0500278 if (inode->i_op->permission)
Nick Pigginb74c79e2011-01-07 17:49:58 +1100279 retval = inode->i_op->permission(inode, mask, 0);
Miklos Szeredif696a362008-07-31 13:41:58 +0200280 else
Nick Pigginb74c79e2011-01-07 17:49:58 +1100281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 if (retval)
285 return retval;
286
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700287 retval = devcgroup_inode_permission(inode, mask);
288 if (retval)
289 return retval;
290
Eric Parisd09ca732010-07-23 11:43:57 -0400291 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800294/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 *
299 * Used to check for read/write/execute permissions on an already opened
300 * file.
301 *
302 * Note:
303 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200304 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800305 */
306int file_permission(struct file *file, int mask)
307{
Al Virof419a2e2008-07-22 00:07:17 -0400308 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311/*
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
321 *
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
327 */
328
329int get_write_access(struct inode * inode)
330{
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
334 return -ETXTBSY;
335 }
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
338
339 return 0;
340}
341
342int deny_write_access(struct file * file)
343{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800344 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
349 return -ETXTBSY;
350 }
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
353
354 return 0;
355}
356
Jan Blunck1d957f92008-02-14 19:34:35 -0800357/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800358 * path_get - get a reference to a path
359 * @path: path to get the reference to
360 *
361 * Given a path increment the reference count to the dentry and the vfsmount.
362 */
363void path_get(struct path *path)
364{
365 mntget(path->mnt);
366 dget(path->dentry);
367}
368EXPORT_SYMBOL(path_get);
369
370/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800371 * path_put - put a reference to a path
372 * @path: path to put the reference to
373 *
374 * Given a path decrement the reference count to the dentry and the vfsmount.
375 */
376void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
Jan Blunck1d957f92008-02-14 19:34:35 -0800378 dput(path->dentry);
379 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
Jan Blunck1d957f92008-02-14 19:34:35 -0800381EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
Trond Myklebust834f2a42005-10-18 14:20:16 -0700383/**
Nick Piggin31e6b012011-01-07 17:49:52 +1100384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800386 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100387 *
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
396 *
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
399 */
400static int nameidata_drop_rcu(struct nameidata *nd)
401{
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
404
405 BUG_ON(!(nd->flags & LOOKUP_RCU));
406 if (nd->root.mnt) {
407 spin_lock(&fs->lock);
408 if (nd->root.mnt != fs->root.mnt ||
409 nd->root.dentry != fs->root.dentry)
410 goto err_root;
411 }
412 spin_lock(&dentry->d_lock);
413 if (!__d_rcu_to_refcount(dentry, nd->seq))
414 goto err;
415 BUG_ON(nd->inode != dentry->d_inode);
416 spin_unlock(&dentry->d_lock);
417 if (nd->root.mnt) {
418 path_get(&nd->root);
419 spin_unlock(&fs->lock);
420 }
421 mntget(nd->path.mnt);
422
423 rcu_read_unlock();
424 br_read_unlock(vfsmount_lock);
425 nd->flags &= ~LOOKUP_RCU;
426 return 0;
427err:
428 spin_unlock(&dentry->d_lock);
429err_root:
430 if (nd->root.mnt)
431 spin_unlock(&fs->lock);
432 return -ECHILD;
433}
434
435/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
437{
438 if (nd->flags & LOOKUP_RCU)
439 return nameidata_drop_rcu(nd);
440 return 0;
441}
442
443/**
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800447 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100448 *
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
452 */
453static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
454{
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
457
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
459 if (nd->root.mnt) {
460 spin_lock(&fs->lock);
461 if (nd->root.mnt != fs->root.mnt ||
462 nd->root.dentry != fs->root.dentry)
463 goto err_root;
464 }
465 spin_lock(&parent->d_lock);
466 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
467 if (!__d_rcu_to_refcount(dentry, nd->seq))
468 goto err;
469 /*
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
473 */
474 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
475 BUG_ON(!parent->d_count);
476 parent->d_count++;
477 spin_unlock(&dentry->d_lock);
478 spin_unlock(&parent->d_lock);
479 if (nd->root.mnt) {
480 path_get(&nd->root);
481 spin_unlock(&fs->lock);
482 }
483 mntget(nd->path.mnt);
484
485 rcu_read_unlock();
486 br_read_unlock(vfsmount_lock);
487 nd->flags &= ~LOOKUP_RCU;
488 return 0;
489err:
490 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
492err_root:
493 if (nd->root.mnt)
494 spin_unlock(&fs->lock);
495 return -ECHILD;
496}
497
498/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
500{
Al Viroa7472ba2011-03-04 14:39:30 -0500501 if (nd->flags & LOOKUP_RCU) {
502 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
503 nd->flags &= ~LOOKUP_RCU;
504 nd->root.mnt = NULL;
505 rcu_read_unlock();
506 br_read_unlock(vfsmount_lock);
507 return -ECHILD;
508 }
509 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100510 return 0;
511}
512
513/**
514 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
515 * @nd: nameidata pathwalk data to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800516 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100517 *
518 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
519 * nd->path should be the final element of the lookup, so nd->root is discarded.
520 * Must be called from rcu-walk context.
521 */
522static int nameidata_drop_rcu_last(struct nameidata *nd)
523{
524 struct dentry *dentry = nd->path.dentry;
525
526 BUG_ON(!(nd->flags & LOOKUP_RCU));
527 nd->flags &= ~LOOKUP_RCU;
528 nd->root.mnt = NULL;
529 spin_lock(&dentry->d_lock);
530 if (!__d_rcu_to_refcount(dentry, nd->seq))
531 goto err_unlock;
532 BUG_ON(nd->inode != dentry->d_inode);
533 spin_unlock(&dentry->d_lock);
534
535 mntget(nd->path.mnt);
536
537 rcu_read_unlock();
538 br_read_unlock(vfsmount_lock);
539
540 return 0;
541
542err_unlock:
543 spin_unlock(&dentry->d_lock);
544 rcu_read_unlock();
545 br_read_unlock(vfsmount_lock);
546 return -ECHILD;
547}
548
Nick Piggin31e6b012011-01-07 17:49:52 +1100549/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700550 * release_open_intent - free up open intent resources
551 * @nd: pointer to nameidata
552 */
553void release_open_intent(struct nameidata *nd)
554{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800555 struct file *file = nd->intent.open.file;
556
557 if (file && !IS_ERR(file)) {
558 if (file->f_path.dentry == NULL)
559 put_filp(file);
560 else
561 fput(file);
562 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700563}
564
Al Virof60aef72011-02-15 01:35:28 -0500565static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100566{
Al Virof60aef72011-02-15 01:35:28 -0500567 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100568}
569
Al Virof5e1c1c2011-02-15 01:32:55 -0500570static struct dentry *
Ian Kentbcdc5e02006-09-27 01:50:44 -0700571do_revalidate(struct dentry *dentry, struct nameidata *nd)
572{
Al Virof5e1c1c2011-02-15 01:32:55 -0500573 int status = d_revalidate(dentry, nd);
Ian Kentbcdc5e02006-09-27 01:50:44 -0700574 if (unlikely(status <= 0)) {
575 /*
576 * The dentry failed validation.
577 * If d_revalidate returned 0 attempt to invalidate
578 * the dentry otherwise d_revalidate is asking us
579 * to return a fail status.
580 */
Nick Piggin34286d62011-01-07 17:49:57 +1100581 if (status < 0) {
Al Virof5e1c1c2011-02-15 01:32:55 -0500582 dput(dentry);
Nick Piggin34286d62011-01-07 17:49:57 +1100583 dentry = ERR_PTR(status);
Al Virof5e1c1c2011-02-15 01:32:55 -0500584 } else if (!d_invalidate(dentry)) {
585 dput(dentry);
586 dentry = NULL;
Ian Kentbcdc5e02006-09-27 01:50:44 -0700587 }
588 }
589 return dentry;
590}
591
Al Virof5e1c1c2011-02-15 01:32:55 -0500592static inline struct dentry *
593do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
594{
Al Virof60aef72011-02-15 01:35:28 -0500595 int status = d_revalidate(dentry, nd);
Al Virof5e1c1c2011-02-15 01:32:55 -0500596 if (likely(status > 0))
597 return dentry;
598 if (status == -ECHILD) {
599 if (nameidata_dentry_drop_rcu(nd, dentry))
600 return ERR_PTR(-ECHILD);
601 return do_revalidate(dentry, nd);
602 }
603 if (status < 0)
604 return ERR_PTR(status);
605 /* Don't d_invalidate in rcu-walk mode */
606 if (nameidata_dentry_drop_rcu(nd, dentry))
607 return ERR_PTR(-ECHILD);
608 if (!d_invalidate(dentry)) {
609 dput(dentry);
610 dentry = NULL;
611 }
612 return dentry;
613}
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615/*
Al Viro16c2cd72011-02-22 15:50:10 -0500616 * handle_reval_path - force revalidation of a dentry
Jeff Layton39159de2009-12-07 12:01:50 -0500617 *
618 * In some situations the path walking code will trust dentries without
619 * revalidating them. This causes problems for filesystems that depend on
620 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
621 * (which indicates that it's possible for the dentry to go stale), force
622 * a d_revalidate call before proceeding.
623 *
624 * Returns 0 if the revalidation was successful. If the revalidation fails,
625 * either return the error returned by d_revalidate or -ESTALE if the
626 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
627 * invalidate the dentry. It's up to the caller to handle putting references
628 * to the path if necessary.
629 */
Al Viro16c2cd72011-02-22 15:50:10 -0500630static inline int handle_reval_path(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500631{
Al Viro16c2cd72011-02-22 15:50:10 -0500632 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500633 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500634
Al Viro16c2cd72011-02-22 15:50:10 -0500635 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500636 return 0;
637
Al Viro16c2cd72011-02-22 15:50:10 -0500638 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
639 return 0;
640
641 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
642 return 0;
643
644 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100645 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500646 if (status > 0)
647 return 0;
648
Al Viro16c2cd72011-02-22 15:50:10 -0500649 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500650 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500651
Jeff Layton39159de2009-12-07 12:01:50 -0500652 return status;
653}
654
655/*
Al Virob75b5082009-12-16 01:01:38 -0500656 * Short-cut version of permission(), for calling on directories
657 * during pathname resolution. Combines parts of permission()
658 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 *
660 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500661 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 * complete permission check.
663 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100664static inline int exec_permission(struct inode *inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700666 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700668 if (inode->i_op->permission) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100669 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
670 } else {
671 ret = acl_permission_check(inode, MAY_EXEC, flags,
672 inode->i_op->check_acl);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700673 }
Nick Pigginb74c79e2011-01-07 17:49:58 +1100674 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 goto ok;
Nick Pigginb74c79e2011-01-07 17:49:58 +1100676 if (ret == -ECHILD)
Nick Piggin31e6b012011-01-07 17:49:52 +1100677 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700679 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 goto ok;
681
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700682 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683ok:
Nick Pigginb74c79e2011-01-07 17:49:58 +1100684 return security_inode_exec_permission(inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
Al Viro2a737872009-04-07 11:49:53 -0400687static __always_inline void set_root(struct nameidata *nd)
688{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200689 if (!nd->root.mnt)
690 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400691}
692
Al Viro6de88d72009-08-09 01:41:57 +0400693static int link_path_walk(const char *, struct nameidata *);
694
Nick Piggin31e6b012011-01-07 17:49:52 +1100695static __always_inline void set_root_rcu(struct nameidata *nd)
696{
697 if (!nd->root.mnt) {
698 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100699 unsigned seq;
700
701 do {
702 seq = read_seqcount_begin(&fs->seq);
703 nd->root = fs->root;
704 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100705 }
706}
707
Arjan van de Venf1662352006-01-14 13:21:31 -0800708static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709{
Nick Piggin31e6b012011-01-07 17:49:52 +1100710 int ret;
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (IS_ERR(link))
713 goto fail;
714
715 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400716 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800717 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400718 nd->path = nd->root;
719 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500720 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100722 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100723
Nick Piggin31e6b012011-01-07 17:49:52 +1100724 ret = link_path_walk(link, nd);
725 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800727 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 return PTR_ERR(link);
729}
730
Jan Blunck1d957f92008-02-14 19:34:35 -0800731static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700732{
733 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800734 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700735 mntput(path->mnt);
736}
737
Nick Piggin7b9337a2011-01-14 08:42:43 +0000738static inline void path_to_nameidata(const struct path *path,
739 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700740{
Nick Piggin31e6b012011-01-07 17:49:52 +1100741 if (!(nd->flags & LOOKUP_RCU)) {
742 dput(nd->path.dentry);
743 if (nd->path.mnt != path->mnt)
744 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800745 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100746 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800747 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700748}
749
Al Virodef4af32009-12-26 08:37:05 -0500750static __always_inline int
Nick Piggin7b9337a2011-01-14 08:42:43 +0000751__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800752{
753 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000754 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800755
Al Viro844a3912011-02-15 00:38:26 -0500756 BUG_ON(nd->flags & LOOKUP_RCU);
757
Nick Piggin7b9337a2011-01-14 08:42:43 +0000758 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800759 nd_set_link(nd, NULL);
760
David Howells87556ef2011-01-14 18:46:46 +0000761 if (link->mnt == nd->path.mnt)
762 mntget(link->mnt);
Nick Piggin31e6b012011-01-07 17:49:52 +1100763
Al Viro36f3b4f2011-02-22 21:24:38 -0500764 error = security_inode_follow_link(link->dentry, nd);
765 if (error) {
766 *p = ERR_PTR(error); /* no ->put_link(), please */
767 path_put(&nd->path);
768 return error;
769 }
770
Al Viro86acdca12009-12-22 23:45:11 -0500771 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500772 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
773 error = PTR_ERR(*p);
774 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800775 char *s = nd_get_link(nd);
776 error = 0;
777 if (s)
778 error = __vfs_follow_link(nd, s);
Al Viro16c2cd72011-02-22 15:50:10 -0500779 else if (nd->last_type == LAST_BIND)
780 nd->flags |= LOOKUP_JUMPED;
Ian Kent051d3812006-03-27 01:14:53 -0800781 }
Ian Kent051d3812006-03-27 01:14:53 -0800782 return error;
783}
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785/*
786 * This limits recursive symlink follows to 8, while
787 * limiting consecutive symlinks to 40.
788 *
789 * Without that kind of total limit, nasty chains of consecutive
790 * symlinks can cause almost arbitrarily long lookups.
791 */
Linus Torvalds3abb17e2011-02-16 08:56:55 -0800792static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Al Virodef4af32009-12-26 08:37:05 -0500794 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 int err = -ELOOP;
Al Viro844a3912011-02-15 00:38:26 -0500796
797 /* We drop rcu-walk here */
798 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
799 return -ECHILD;
Linus Torvalds3abb17e2011-02-16 08:56:55 -0800800 BUG_ON(inode != path->dentry->d_inode);
Al Viro844a3912011-02-15 00:38:26 -0500801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if (current->link_count >= MAX_NESTED_LINKS)
803 goto loop;
804 if (current->total_link_count >= 40)
805 goto loop;
806 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
807 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 current->link_count++;
809 current->total_link_count++;
810 nd->depth++;
Al Virodef4af32009-12-26 08:37:05 -0500811 err = __do_follow_link(path, nd, &cookie);
812 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
813 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
Al Viro258fa992009-08-09 01:32:02 +0400814 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 current->link_count--;
816 nd->depth--;
817 return err;
818loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800819 path_put_conditional(path, nd);
820 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 return err;
822}
823
Nick Piggin31e6b012011-01-07 17:49:52 +1100824static int follow_up_rcu(struct path *path)
825{
826 struct vfsmount *parent;
827 struct dentry *mountpoint;
828
829 parent = path->mnt->mnt_parent;
830 if (parent == path->mnt)
831 return 0;
832 mountpoint = path->mnt->mnt_mountpoint;
833 path->dentry = mountpoint;
834 path->mnt = parent;
835 return 1;
836}
837
Al Virobab77eb2009-04-18 03:26:48 -0400838int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
840 struct vfsmount *parent;
841 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000842
843 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400844 parent = path->mnt->mnt_parent;
845 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000846 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return 0;
848 }
849 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400850 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000851 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400852 dput(path->dentry);
853 path->dentry = mountpoint;
854 mntput(path->mnt);
855 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 return 1;
857}
858
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100859/*
David Howells9875cf82011-01-14 18:45:21 +0000860 * Perform an automount
861 * - return -EISDIR to tell follow_managed() to stop and return the path we
862 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 */
David Howells9875cf82011-01-14 18:45:21 +0000864static int follow_automount(struct path *path, unsigned flags,
865 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100866{
David Howells9875cf82011-01-14 18:45:21 +0000867 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000868 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100869
David Howells9875cf82011-01-14 18:45:21 +0000870 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
871 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700872
David Howells6f45b652011-01-14 18:45:31 +0000873 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
874 * and this is the terminal part of the path.
875 */
876 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
877 return -EISDIR; /* we actually want to stop here */
878
David Howells9875cf82011-01-14 18:45:21 +0000879 /* We want to mount if someone is trying to open/create a file of any
880 * type under the mountpoint, wants to traverse through the mountpoint
881 * or wants to open the mounted directory.
882 *
883 * We don't want to mount if someone's just doing a stat and they've
884 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
885 * appended a '/' to the name.
886 */
887 if (!(flags & LOOKUP_FOLLOW) &&
888 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
889 LOOKUP_OPEN | LOOKUP_CREATE)))
890 return -EISDIR;
891
892 current->total_link_count++;
893 if (current->total_link_count >= 40)
894 return -ELOOP;
895
896 mnt = path->dentry->d_op->d_automount(path);
897 if (IS_ERR(mnt)) {
898 /*
899 * The filesystem is allowed to return -EISDIR here to indicate
900 * it doesn't want to automount. For instance, autofs would do
901 * this so that its userspace daemon can mount on this dentry.
902 *
903 * However, we can only permit this if it's a terminal point in
904 * the path being looked up; if it wasn't then the remainder of
905 * the path is inaccessible and we should say so.
906 */
907 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
908 return -EREMOTE;
909 return PTR_ERR(mnt);
910 }
David Howellsea5b7782011-01-14 19:10:03 +0000911
David Howells9875cf82011-01-14 18:45:21 +0000912 if (!mnt) /* mount collision */
913 return 0;
914
Al Viro19a167a2011-01-17 01:35:23 -0500915 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000916
David Howellsea5b7782011-01-14 19:10:03 +0000917 switch (err) {
918 case -EBUSY:
919 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500920 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000921 case 0:
David Howellsea5b7782011-01-14 19:10:03 +0000922 dput(path->dentry);
923 if (*need_mntput)
924 mntput(path->mnt);
925 path->mnt = mnt;
926 path->dentry = dget(mnt->mnt_root);
927 *need_mntput = true;
928 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500929 default:
930 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000931 }
Al Viro19a167a2011-01-17 01:35:23 -0500932
David Howells9875cf82011-01-14 18:45:21 +0000933}
934
935/*
936 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000937 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000938 * - Flagged as mountpoint
939 * - Flagged as automount point
940 *
941 * This may only be called in refwalk mode.
942 *
943 * Serialization is taken care of in namespace.c
944 */
945static int follow_managed(struct path *path, unsigned flags)
946{
947 unsigned managed;
948 bool need_mntput = false;
949 int ret;
950
951 /* Given that we're not holding a lock here, we retain the value in a
952 * local variable for each dentry as we look at it so that we don't see
953 * the components of that value change under us */
954 while (managed = ACCESS_ONCE(path->dentry->d_flags),
955 managed &= DCACHE_MANAGED_DENTRY,
956 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000957 /* Allow the filesystem to manage the transit without i_mutex
958 * being held. */
959 if (managed & DCACHE_MANAGE_TRANSIT) {
960 BUG_ON(!path->dentry->d_op);
961 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000962 ret = path->dentry->d_op->d_manage(path->dentry,
963 false, false);
David Howellscc53ce52011-01-14 18:45:26 +0000964 if (ret < 0)
965 return ret == -EISDIR ? 0 : ret;
966 }
967
David Howells9875cf82011-01-14 18:45:21 +0000968 /* Transit to a mounted filesystem. */
969 if (managed & DCACHE_MOUNTED) {
970 struct vfsmount *mounted = lookup_mnt(path);
971 if (mounted) {
972 dput(path->dentry);
973 if (need_mntput)
974 mntput(path->mnt);
975 path->mnt = mounted;
976 path->dentry = dget(mounted->mnt_root);
977 need_mntput = true;
978 continue;
979 }
980
981 /* Something is mounted on this dentry in another
982 * namespace and/or whatever was mounted there in this
983 * namespace got unmounted before we managed to get the
984 * vfsmount_lock */
985 }
986
987 /* Handle an automount point */
988 if (managed & DCACHE_NEED_AUTOMOUNT) {
989 ret = follow_automount(path, flags, &need_mntput);
990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 continue;
993 }
994
995 /* We didn't change the current path point */
996 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 }
David Howells9875cf82011-01-14 18:45:21 +0000998 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999}
1000
David Howellscc53ce52011-01-14 18:45:26 +00001001int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002{
1003 struct vfsmount *mounted;
1004
Al Viro1c755af2009-04-18 14:06:57 -04001005 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001007 dput(path->dentry);
1008 mntput(path->mnt);
1009 path->mnt = mounted;
1010 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 return 1;
1012 }
1013 return 0;
1014}
1015
David Howells9875cf82011-01-14 18:45:21 +00001016/*
1017 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
David Howellscc53ce52011-01-14 18:45:26 +00001018 * meet a managed dentry and we're not walking to "..". True is returned to
David Howells9875cf82011-01-14 18:45:21 +00001019 * continue, false to abort.
1020 */
1021static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1022 struct inode **inode, bool reverse_transit)
1023{
1024 while (d_mountpoint(path->dentry)) {
1025 struct vfsmount *mounted;
David Howellsab909112011-01-14 18:46:51 +00001026 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1027 !reverse_transit &&
1028 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1029 return false;
David Howells9875cf82011-01-14 18:45:21 +00001030 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1031 if (!mounted)
1032 break;
1033 path->mnt = mounted;
1034 path->dentry = mounted->mnt_root;
1035 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1036 *inode = path->dentry->d_inode;
1037 }
1038
1039 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1040 return reverse_transit;
1041 return true;
1042}
1043
Nick Piggin31e6b012011-01-07 17:49:52 +11001044static int follow_dotdot_rcu(struct nameidata *nd)
1045{
1046 struct inode *inode = nd->inode;
1047
1048 set_root_rcu(nd);
1049
David Howells9875cf82011-01-14 18:45:21 +00001050 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001051 if (nd->path.dentry == nd->root.dentry &&
1052 nd->path.mnt == nd->root.mnt) {
1053 break;
1054 }
1055 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1056 struct dentry *old = nd->path.dentry;
1057 struct dentry *parent = old->d_parent;
1058 unsigned seq;
1059
1060 seq = read_seqcount_begin(&parent->d_seq);
1061 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001062 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001063 inode = parent->d_inode;
1064 nd->path.dentry = parent;
1065 nd->seq = seq;
1066 break;
1067 }
1068 if (!follow_up_rcu(&nd->path))
1069 break;
1070 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1071 inode = nd->path.dentry->d_inode;
1072 }
David Howells9875cf82011-01-14 18:45:21 +00001073 __follow_mount_rcu(nd, &nd->path, &inode, true);
Nick Piggin31e6b012011-01-07 17:49:52 +11001074 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001075 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001076
1077failed:
1078 nd->flags &= ~LOOKUP_RCU;
1079 nd->root.mnt = NULL;
1080 rcu_read_unlock();
1081 br_read_unlock(vfsmount_lock);
1082 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001083}
1084
David Howells9875cf82011-01-14 18:45:21 +00001085/*
David Howellscc53ce52011-01-14 18:45:26 +00001086 * Follow down to the covering mount currently visible to userspace. At each
1087 * point, the filesystem owning that dentry may be queried as to whether the
1088 * caller is permitted to proceed or not.
1089 *
1090 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1091 * being true).
1092 */
1093int follow_down(struct path *path, bool mounting_here)
1094{
1095 unsigned managed;
1096 int ret;
1097
1098 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1099 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1100 /* Allow the filesystem to manage the transit without i_mutex
1101 * being held.
1102 *
1103 * We indicate to the filesystem if someone is trying to mount
1104 * something here. This gives autofs the chance to deny anyone
1105 * other than its daemon the right to mount on its
1106 * superstructure.
1107 *
1108 * The filesystem may sleep at this point.
1109 */
1110 if (managed & DCACHE_MANAGE_TRANSIT) {
1111 BUG_ON(!path->dentry->d_op);
1112 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001113 ret = path->dentry->d_op->d_manage(
1114 path->dentry, mounting_here, false);
David Howellscc53ce52011-01-14 18:45:26 +00001115 if (ret < 0)
1116 return ret == -EISDIR ? 0 : ret;
1117 }
1118
1119 /* Transit to a mounted filesystem. */
1120 if (managed & DCACHE_MOUNTED) {
1121 struct vfsmount *mounted = lookup_mnt(path);
1122 if (!mounted)
1123 break;
1124 dput(path->dentry);
1125 mntput(path->mnt);
1126 path->mnt = mounted;
1127 path->dentry = dget(mounted->mnt_root);
1128 continue;
1129 }
1130
1131 /* Don't handle automount points here */
1132 break;
1133 }
1134 return 0;
1135}
1136
1137/*
David Howells9875cf82011-01-14 18:45:21 +00001138 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1139 */
1140static void follow_mount(struct path *path)
1141{
1142 while (d_mountpoint(path->dentry)) {
1143 struct vfsmount *mounted = lookup_mnt(path);
1144 if (!mounted)
1145 break;
1146 dput(path->dentry);
1147 mntput(path->mnt);
1148 path->mnt = mounted;
1149 path->dentry = dget(mounted->mnt_root);
1150 }
1151}
1152
Nick Piggin31e6b012011-01-07 17:49:52 +11001153static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154{
Al Viro2a737872009-04-07 11:49:53 -04001155 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001158 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Al Viro2a737872009-04-07 11:49:53 -04001160 if (nd->path.dentry == nd->root.dentry &&
1161 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 break;
1163 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001164 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001165 /* rare case of legitimate dget_parent()... */
1166 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 dput(old);
1168 break;
1169 }
Al Viro3088dd72010-01-30 15:47:29 -05001170 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 }
Al Viro79ed0222009-04-18 13:59:41 -04001173 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001174 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
1176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001178 * Allocate a dentry with name and parent, and perform a parent
1179 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1180 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1181 * have verified that no child exists while under i_mutex.
1182 */
1183static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1184 struct qstr *name, struct nameidata *nd)
1185{
1186 struct inode *inode = parent->d_inode;
1187 struct dentry *dentry;
1188 struct dentry *old;
1189
1190 /* Don't create child dentry for a dead directory. */
1191 if (unlikely(IS_DEADDIR(inode)))
1192 return ERR_PTR(-ENOENT);
1193
1194 dentry = d_alloc(parent, name);
1195 if (unlikely(!dentry))
1196 return ERR_PTR(-ENOMEM);
1197
1198 old = inode->i_op->lookup(inode, dentry, nd);
1199 if (unlikely(old)) {
1200 dput(dentry);
1201 dentry = old;
1202 }
1203 return dentry;
1204}
1205
1206/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 * It's more convoluted than I'd like it to be, but... it's still fairly
1208 * small and for now I'd prefer to have fast path as straight as possible.
1209 * It _is_ time-critical.
1210 */
1211static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001212 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
Jan Blunck4ac91372008-02-14 19:34:32 -08001214 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001215 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro6e6b1bd2009-08-13 23:38:37 +04001216 struct inode *dir;
David Howells9875cf82011-01-14 18:45:21 +00001217 int err;
1218
Al Viro3cac2602009-08-13 18:27:43 +04001219 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001220 * Rename seqlock is not required here because in the off chance
1221 * of a false negative due to a concurrent rename, we're going to
1222 * do the non-racy lookup, below.
1223 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001224 if (nd->flags & LOOKUP_RCU) {
1225 unsigned seq;
1226
1227 *inode = nd->inode;
1228 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1229 if (!dentry) {
1230 if (nameidata_drop_rcu(nd))
1231 return -ECHILD;
1232 goto need_lookup;
1233 }
1234 /* Memory barrier in read_seqcount_begin of child is enough */
1235 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1236 return -ECHILD;
1237
1238 nd->seq = seq;
Al Viro24643082011-02-15 01:26:22 -05001239 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Virof5e1c1c2011-02-15 01:32:55 -05001240 dentry = do_revalidate_rcu(dentry, nd);
Al Viro24643082011-02-15 01:26:22 -05001241 if (!dentry)
1242 goto need_lookup;
1243 if (IS_ERR(dentry))
1244 goto fail;
1245 if (!(nd->flags & LOOKUP_RCU))
1246 goto done;
1247 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001248 path->mnt = mnt;
1249 path->dentry = dentry;
David Howells9875cf82011-01-14 18:45:21 +00001250 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1251 return 0;
1252 if (nameidata_drop_rcu(nd))
1253 return -ECHILD;
1254 /* fallthru */
Nick Piggin31e6b012011-01-07 17:49:52 +11001255 }
David Howells9875cf82011-01-14 18:45:21 +00001256 dentry = __d_lookup(parent, name);
1257 if (!dentry)
1258 goto need_lookup;
1259found:
Al Viro24643082011-02-15 01:26:22 -05001260 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1261 dentry = do_revalidate(dentry, nd);
1262 if (!dentry)
1263 goto need_lookup;
1264 if (IS_ERR(dentry))
1265 goto fail;
1266 }
David Howells9875cf82011-01-14 18:45:21 +00001267done:
1268 path->mnt = mnt;
1269 path->dentry = dentry;
1270 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001271 if (unlikely(err < 0)) {
1272 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001273 return err;
Ian Kent89312212011-01-18 12:06:10 +08001274 }
David Howells9875cf82011-01-14 18:45:21 +00001275 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 return 0;
1277
1278need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +04001279 dir = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001280 BUG_ON(nd->inode != dir);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001281
1282 mutex_lock(&dir->i_mutex);
1283 /*
1284 * First re-do the cached lookup just in case it was created
Nick Pigginb04f7842010-08-18 04:37:34 +10001285 * while we waited for the directory semaphore, or the first
1286 * lookup failed due to an unrelated rename.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001287 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001288 * This could use version numbering or similar to avoid unnecessary
1289 * cache lookups, but then we'd have to do the first lookup in the
1290 * non-racy way. However in the common case here, everything should
1291 * be hot in cache, so would it be a big win?
Al Viro6e6b1bd2009-08-13 23:38:37 +04001292 */
1293 dentry = d_lookup(parent, name);
Nick Pigginbaa03892010-08-18 04:37:31 +10001294 if (likely(!dentry)) {
1295 dentry = d_alloc_and_lookup(parent, name, nd);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001296 mutex_unlock(&dir->i_mutex);
1297 if (IS_ERR(dentry))
1298 goto fail;
1299 goto done;
1300 }
Al Viro6e6b1bd2009-08-13 23:38:37 +04001301 /*
1302 * Uhhuh! Nasty case: the cache was re-populated while
1303 * we waited on the semaphore. Need to revalidate.
1304 */
1305 mutex_unlock(&dir->i_mutex);
Nick Piggin2e2e88e2010-08-18 04:37:30 +10001306 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308fail:
1309 return PTR_ERR(dentry);
1310}
1311
Al Viro52094c82011-02-21 21:34:47 -05001312static inline int may_lookup(struct nameidata *nd)
1313{
1314 if (nd->flags & LOOKUP_RCU) {
1315 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1316 if (err != -ECHILD)
1317 return err;
1318 if (nameidata_drop_rcu(nd))
1319 return -ECHILD;
1320 }
1321 return exec_permission(nd->inode, 0);
1322}
1323
Al Viro9856fa12011-03-04 14:22:06 -05001324static inline int handle_dots(struct nameidata *nd, int type)
1325{
1326 if (type == LAST_DOTDOT) {
1327 if (nd->flags & LOOKUP_RCU) {
1328 if (follow_dotdot_rcu(nd))
1329 return -ECHILD;
1330 } else
1331 follow_dotdot(nd);
1332 }
1333 return 0;
1334}
1335
Al Viro951361f2011-03-04 14:44:37 -05001336static void terminate_walk(struct nameidata *nd)
1337{
1338 if (!(nd->flags & LOOKUP_RCU)) {
1339 path_put(&nd->path);
1340 } else {
1341 nd->flags &= ~LOOKUP_RCU;
1342 nd->root.mnt = NULL;
1343 rcu_read_unlock();
1344 br_read_unlock(vfsmount_lock);
1345 }
1346}
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348/*
1349 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001350 * This is the basic name resolution function, turning a pathname into
1351 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001353 * Returns 0 and nd will have valid dentry and mnt on success.
1354 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 */
Al Viro6de88d72009-08-09 01:41:57 +04001356static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
1358 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 int err;
1360 unsigned int lookup_flags = nd->flags;
1361
1362 while (*name=='/')
1363 name++;
1364 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001365 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -08001368 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 /* At this point we know we have a real path component. */
1371 for(;;) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001372 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 unsigned long hash;
1374 struct qstr this;
1375 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001376 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Trond Myklebustcdce5d62005-10-18 14:20:18 -07001378 nd->flags |= LOOKUP_CONTINUE;
Al Viro52094c82011-02-21 21:34:47 -05001379
1380 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 if (err)
1382 break;
1383
1384 this.name = name;
1385 c = *(const unsigned char *)name;
1386
1387 hash = init_name_hash();
1388 do {
1389 name++;
1390 hash = partial_name_hash(c, hash);
1391 c = *(const unsigned char *)name;
1392 } while (c && (c != '/'));
1393 this.len = name - (const char *) this.name;
1394 this.hash = end_name_hash(hash);
1395
Al Virofe479a52011-02-22 15:10:03 -05001396 type = LAST_NORM;
1397 if (this.name[0] == '.') switch (this.len) {
1398 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001399 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001400 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001401 nd->flags |= LOOKUP_JUMPED;
1402 }
Al Virofe479a52011-02-22 15:10:03 -05001403 break;
1404 case 1:
1405 type = LAST_DOT;
1406 }
Al Viro5a202bc2011-03-08 14:17:44 -05001407 if (likely(type == LAST_NORM)) {
1408 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001409 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001410 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1411 err = parent->d_op->d_hash(parent, nd->inode,
1412 &this);
1413 if (err < 0)
1414 break;
1415 }
1416 }
Al Virofe479a52011-02-22 15:10:03 -05001417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 /* remove trailing slashes? */
1419 if (!c)
1420 goto last_component;
1421 while (*++name == '/');
1422 if (!*name)
1423 goto last_with_slashes;
1424
1425 /*
1426 * "." and ".." are special - ".." especially so because it has
1427 * to be able to know about the current root directory and
1428 * parent relationships.
1429 */
Al Virofe479a52011-02-22 15:10:03 -05001430 if (unlikely(type != LAST_NORM)) {
Al Viroef7562d2011-03-04 14:35:59 -05001431 if (handle_dots(nd, type))
1432 return -ECHILD;
Al Virofe479a52011-02-22 15:10:03 -05001433 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 }
Al Virofe479a52011-02-22 15:10:03 -05001435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 /* This does the actual lookups.. */
Nick Piggin31e6b012011-01-07 17:49:52 +11001437 err = do_lookup(nd, &this, &next, &inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 if (err)
1439 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Al Viro7bc055d2011-02-23 19:41:31 -05001441 if (inode && inode->i_op->follow_link) {
Linus Torvalds3abb17e2011-02-16 08:56:55 -08001442 err = do_follow_link(inode, &next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001444 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001445 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001446 } else {
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001447 path_to_nameidata(&next, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001448 nd->inode = inode;
1449 }
Al Viro7bc055d2011-02-23 19:41:31 -05001450 err = -ENOENT;
1451 if (!nd->inode)
1452 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001454 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 break;
1456 continue;
1457 /* here ends the main loop */
1458
1459last_with_slashes:
1460 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1461last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -08001462 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1463 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (lookup_flags & LOOKUP_PARENT)
1465 goto lookup_parent;
Al Viroef7562d2011-03-04 14:35:59 -05001466 if (unlikely(type != LAST_NORM))
1467 return handle_dots(nd, type);
Nick Piggin31e6b012011-01-07 17:49:52 +11001468 err = do_lookup(nd, &this, &next, &inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (err)
1470 break;
David Howellsdb372912011-01-14 18:45:53 +00001471 if (inode && unlikely(inode->i_op->follow_link) &&
1472 (lookup_flags & LOOKUP_FOLLOW)) {
Linus Torvalds3abb17e2011-02-16 08:56:55 -08001473 err = do_follow_link(inode, &next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001475 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001476 nd->inode = nd->path.dentry->d_inode;
1477 } else {
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001478 path_to_nameidata(&next, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001479 nd->inode = inode;
1480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 err = -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001482 if (!nd->inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 break;
1484 if (lookup_flags & LOOKUP_DIRECTORY) {
1485 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001486 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 break;
1488 }
Al Viro086e1832011-02-22 20:56:27 -05001489 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490lookup_parent:
1491 nd->last = this;
Al Virofe479a52011-02-22 15:10:03 -05001492 nd->last_type = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
Al Viro951361f2011-03-04 14:44:37 -05001495 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 return err;
1497}
1498
Al Viro70e9b352011-03-05 21:12:22 -05001499static int path_init(int dfd, const char *name, unsigned int flags,
1500 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001502 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001503 int fput_needed;
1504 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001507 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001509 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001512 if (flags & LOOKUP_RCU) {
1513 br_read_lock(vfsmount_lock);
1514 rcu_read_lock();
1515 set_root_rcu(nd);
1516 } else {
1517 set_root(nd);
1518 path_get(&nd->root);
1519 }
Al Viro2a737872009-04-07 11:49:53 -04001520 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001521 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001522 if (flags & LOOKUP_RCU) {
1523 struct fs_struct *fs = current->fs;
1524 unsigned seq;
1525
1526 br_read_lock(vfsmount_lock);
1527 rcu_read_lock();
1528
1529 do {
1530 seq = read_seqcount_begin(&fs->seq);
1531 nd->path = fs->pwd;
1532 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1533 } while (read_seqcount_retry(&fs->seq, seq));
1534 } else {
1535 get_fs_pwd(current->fs, &nd->path);
1536 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001537 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001538 struct dentry *dentry;
1539
1540 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001541 retval = -EBADF;
1542 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001543 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001544
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001545 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001546
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001547 retval = -ENOTDIR;
1548 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001549 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001550
1551 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001552 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001553 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001554
Jan Blunck5dd784d2008-02-14 19:34:38 -08001555 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001556 if (flags & LOOKUP_RCU) {
1557 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001558 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001559 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1560 br_read_lock(vfsmount_lock);
1561 rcu_read_lock();
1562 } else {
1563 path_get(&file->f_path);
1564 fput_light(file, fput_needed);
1565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 }
Al Viroe41f7d42011-02-22 14:02:58 -05001567
Nick Piggin31e6b012011-01-07 17:49:52 +11001568 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001569 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001570
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001571fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001572 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001573out_fail:
1574 return retval;
1575}
1576
1577/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001578static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001579 unsigned int flags, struct nameidata *nd)
1580{
Al Viro70e9b352011-03-05 21:12:22 -05001581 struct file *base = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +11001582 int retval;
1583
1584 /*
1585 * Path walking is largely split up into 2 different synchronisation
1586 * schemes, rcu-walk and ref-walk (explained in
1587 * Documentation/filesystems/path-lookup.txt). These share much of the
1588 * path walk code, but some things particularly setup, cleanup, and
1589 * following mounts are sufficiently divergent that functions are
1590 * duplicated. Typically there is a function foo(), and its RCU
1591 * analogue, foo_rcu().
1592 *
1593 * -ECHILD is the error number of choice (just to avoid clashes) that
1594 * is returned if some aspect of an rcu-walk fails. Such an error must
1595 * be handled by restarting a traditional ref-walk (which will always
1596 * be able to complete).
1597 */
Al Viro70e9b352011-03-05 21:12:22 -05001598 retval = path_init(dfd, name, flags, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001599
Nick Piggin31e6b012011-01-07 17:49:52 +11001600 if (unlikely(retval))
1601 return retval;
Al Viroee0827c2011-02-21 23:38:09 -05001602
1603 current->total_link_count = 0;
1604 retval = link_path_walk(name, nd);
1605
1606 if (nd->flags & LOOKUP_RCU) {
Al Viro4455ca62011-03-04 14:28:10 -05001607 /* went all way through without dropping RCU */
1608 BUG_ON(retval);
1609 if (nameidata_drop_rcu_last(nd))
1610 retval = -ECHILD;
Al Viroee0827c2011-02-21 23:38:09 -05001611 }
1612
Al Viro16c2cd72011-02-22 15:50:10 -05001613 if (!retval)
1614 retval = handle_reval_path(nd);
1615
Al Viro70e9b352011-03-05 21:12:22 -05001616 if (base)
1617 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001618
Al Viro2a737872009-04-07 11:49:53 -04001619 if (nd->root.mnt) {
1620 path_put(&nd->root);
1621 nd->root.mnt = NULL;
1622 }
Al Viroee0827c2011-02-21 23:38:09 -05001623 return retval;
1624}
Nick Piggin31e6b012011-01-07 17:49:52 +11001625
Al Viroee0827c2011-02-21 23:38:09 -05001626static int do_path_lookup(int dfd, const char *name,
1627 unsigned int flags, struct nameidata *nd)
1628{
1629 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1630 if (unlikely(retval == -ECHILD))
1631 retval = path_lookupat(dfd, name, flags, nd);
1632 if (unlikely(retval == -ESTALE))
1633 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001634
1635 if (likely(!retval)) {
1636 if (unlikely(!audit_dummy_context())) {
1637 if (nd->path.dentry && nd->inode)
1638 audit_inode(name, nd->path.dentry);
1639 }
1640 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001641 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
Al Viroc9c6cac2011-02-16 15:15:47 -05001644int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001645{
Al Viroc9c6cac2011-02-16 15:15:47 -05001646 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001647}
1648
Al Virod1811462008-08-02 00:49:18 -04001649int kern_path(const char *name, unsigned int flags, struct path *path)
1650{
1651 struct nameidata nd;
1652 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1653 if (!res)
1654 *path = nd.path;
1655 return res;
1656}
1657
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001658/**
1659 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1660 * @dentry: pointer to dentry of the base directory
1661 * @mnt: pointer to vfs mount of the base directory
1662 * @name: pointer to file name
1663 * @flags: lookup flags
1664 * @nd: pointer to nameidata
1665 */
1666int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1667 const char *name, unsigned int flags,
1668 struct nameidata *nd)
1669{
Al Viroee0827c2011-02-21 23:38:09 -05001670 int result;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001671
1672 /* same as do_path_lookup */
1673 nd->last_type = LAST_ROOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001674 nd->flags = flags | LOOKUP_JUMPED;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001675 nd->depth = 0;
1676
Jan Blunckc8e7f442008-06-09 16:40:35 -07001677 nd->path.dentry = dentry;
1678 nd->path.mnt = mnt;
1679 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001680 nd->root = nd->path;
1681 path_get(&nd->root);
Nick Piggin31e6b012011-01-07 17:49:52 +11001682 nd->inode = nd->path.dentry->d_inode;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001683
Al Viroee0827c2011-02-21 23:38:09 -05001684 current->total_link_count = 0;
1685
1686 result = link_path_walk(name, nd);
Al Viro16c2cd72011-02-22 15:50:10 -05001687 if (!result)
1688 result = handle_reval_path(nd);
Al Viroee0827c2011-02-21 23:38:09 -05001689 if (result == -ESTALE) {
1690 /* nd->path had been dropped */
1691 current->total_link_count = 0;
1692 nd->path.dentry = dentry;
1693 nd->path.mnt = mnt;
1694 nd->inode = dentry->d_inode;
1695 path_get(&nd->path);
Al Viro16c2cd72011-02-22 15:50:10 -05001696 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL;
1697
Al Viroee0827c2011-02-21 23:38:09 -05001698 result = link_path_walk(name, nd);
Al Viro16c2cd72011-02-22 15:50:10 -05001699 if (!result)
1700 result = handle_reval_path(nd);
Al Viroee0827c2011-02-21 23:38:09 -05001701 }
1702 if (unlikely(!result && !audit_dummy_context() && nd->path.dentry &&
Nick Piggin31e6b012011-01-07 17:49:52 +11001703 nd->inode))
Jan Blunck4ac91372008-02-14 19:34:32 -08001704 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001705
Al Viro5b857112009-04-07 11:53:49 -04001706 path_put(&nd->root);
1707 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001708
Al Viroee0827c2011-02-21 23:38:09 -05001709 return result;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001710}
1711
Christoph Hellwigeead1912007-10-16 23:25:38 -07001712static struct dentry *__lookup_hash(struct qstr *name,
1713 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001715 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001716 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 int err;
1718
Nick Pigginb74c79e2011-01-07 17:49:58 +11001719 err = exec_permission(inode, 0);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001720 if (err)
1721 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
1723 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001724 * Don't bother with __d_lookup: callers are for creat as
1725 * well as unlink, so a lot of the time it would cost
1726 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001727 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001728 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001729
Nick Pigginfb045ad2011-01-07 17:49:55 +11001730 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro6e6b1bd2009-08-13 23:38:37 +04001731 dentry = do_revalidate(dentry, nd);
1732
Nick Pigginbaa03892010-08-18 04:37:31 +10001733 if (!dentry)
1734 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001735
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 return dentry;
1737}
1738
James Morris057f6c02007-04-26 00:12:05 -07001739/*
1740 * Restricted form of lookup. Doesn't follow links, single-component only,
1741 * needs parent already locked. Doesn't follow mounts.
1742 * SMP-safe.
1743 */
Adrian Bunka244e162006-03-31 02:32:11 -08001744static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745{
Jan Blunck4ac91372008-02-14 19:34:32 -08001746 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747}
1748
Christoph Hellwigeead1912007-10-16 23:25:38 -07001749/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001750 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001751 * @name: pathname component to lookup
1752 * @base: base directory to lookup from
1753 * @len: maximum length @len should be interpreted to
1754 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001755 * Note that this routine is purely a helper for filesystem usage and should
1756 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001757 * nameidata argument is passed to the filesystem methods and a filesystem
1758 * using this helper needs to be prepared for that.
1759 */
James Morris057f6c02007-04-26 00:12:05 -07001760struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1761{
James Morris057f6c02007-04-26 00:12:05 -07001762 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001763 unsigned long hash;
1764 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001765
David Woodhouse2f9092e2009-04-20 23:18:37 +01001766 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1767
Al Viro6a96ba52011-03-07 23:49:20 -05001768 this.name = name;
1769 this.len = len;
1770 if (!len)
1771 return ERR_PTR(-EACCES);
1772
1773 hash = init_name_hash();
1774 while (len--) {
1775 c = *(const unsigned char *)name++;
1776 if (c == '/' || c == '\0')
1777 return ERR_PTR(-EACCES);
1778 hash = partial_name_hash(c, hash);
1779 }
1780 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001781 /*
1782 * See if the low-level filesystem might want
1783 * to use its own hash..
1784 */
1785 if (base->d_flags & DCACHE_OP_HASH) {
1786 int err = base->d_op->d_hash(base, base->d_inode, &this);
1787 if (err < 0)
1788 return ERR_PTR(err);
1789 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001790
Christoph Hellwig49705b72005-11-08 21:35:06 -08001791 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001792}
1793
Al Viro2d8f3032008-07-22 09:59:21 -04001794int user_path_at(int dfd, const char __user *name, unsigned flags,
1795 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796{
Al Viro2d8f3032008-07-22 09:59:21 -04001797 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 char *tmp = getname(name);
1799 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001801
1802 BUG_ON(flags & LOOKUP_PARENT);
1803
1804 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001806 if (!err)
1807 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
1809 return err;
1810}
1811
Al Viro2ad94ae2008-07-21 09:32:51 -04001812static int user_path_parent(int dfd, const char __user *path,
1813 struct nameidata *nd, char **name)
1814{
1815 char *s = getname(path);
1816 int error;
1817
1818 if (IS_ERR(s))
1819 return PTR_ERR(s);
1820
1821 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1822 if (error)
1823 putname(s);
1824 else
1825 *name = s;
1826
1827 return error;
1828}
1829
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830/*
1831 * It's inline, so penalty for filesystems that don't use sticky bit is
1832 * minimal.
1833 */
1834static inline int check_sticky(struct inode *dir, struct inode *inode)
1835{
David Howellsda9592e2008-11-14 10:39:05 +11001836 uid_t fsuid = current_fsuid();
1837
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 if (!(dir->i_mode & S_ISVTX))
1839 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001840 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001842 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 return 0;
1844 return !capable(CAP_FOWNER);
1845}
1846
1847/*
1848 * Check whether we can remove a link victim from directory dir, check
1849 * whether the type of victim is right.
1850 * 1. We can't do it if dir is read-only (done in permission())
1851 * 2. We should have write and exec permissions on dir
1852 * 3. We can't remove anything from append-only dir
1853 * 4. We can't do anything with immutable dir (done in permission())
1854 * 5. If the sticky bit on dir is set we should either
1855 * a. be owner of dir, or
1856 * b. be owner of victim, or
1857 * c. have CAP_FOWNER capability
1858 * 6. If the victim is append-only or immutable we can't do antyhing with
1859 * links pointing to it.
1860 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1861 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1862 * 9. We can't remove a root or mountpoint.
1863 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1864 * nfs_async_unlink().
1865 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001866static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867{
1868 int error;
1869
1870 if (!victim->d_inode)
1871 return -ENOENT;
1872
1873 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001874 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Al Virof419a2e2008-07-22 00:07:17 -04001876 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (error)
1878 return error;
1879 if (IS_APPEND(dir))
1880 return -EPERM;
1881 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001882 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 return -EPERM;
1884 if (isdir) {
1885 if (!S_ISDIR(victim->d_inode->i_mode))
1886 return -ENOTDIR;
1887 if (IS_ROOT(victim))
1888 return -EBUSY;
1889 } else if (S_ISDIR(victim->d_inode->i_mode))
1890 return -EISDIR;
1891 if (IS_DEADDIR(dir))
1892 return -ENOENT;
1893 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1894 return -EBUSY;
1895 return 0;
1896}
1897
1898/* Check whether we can create an object with dentry child in directory
1899 * dir.
1900 * 1. We can't do it if child already exists (open has special treatment for
1901 * this case, but since we are inlined it's OK)
1902 * 2. We can't do it if dir is read-only (done in permission())
1903 * 3. We should have write and exec permissions on dir
1904 * 4. We can't do it if dir is immutable (done in permission())
1905 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001906static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
1908 if (child->d_inode)
1909 return -EEXIST;
1910 if (IS_DEADDIR(dir))
1911 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001912 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915/*
1916 * p1 and p2 should be directories on the same fs.
1917 */
1918struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1919{
1920 struct dentry *p;
1921
1922 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001923 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 return NULL;
1925 }
1926
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001927 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001929 p = d_ancestor(p2, p1);
1930 if (p) {
1931 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1932 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1933 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 }
1935
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001936 p = d_ancestor(p1, p2);
1937 if (p) {
1938 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1939 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1940 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 }
1942
Ingo Molnarf2eace22006-07-03 00:25:05 -07001943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1944 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 return NULL;
1946}
1947
1948void unlock_rename(struct dentry *p1, struct dentry *p2)
1949{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001950 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001952 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001953 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 }
1955}
1956
1957int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1958 struct nameidata *nd)
1959{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001960 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
1962 if (error)
1963 return error;
1964
Al Viroacfa4382008-12-04 10:06:33 -05001965 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 return -EACCES; /* shouldn't it be ENOSYS? */
1967 mode &= S_IALLUGO;
1968 mode |= S_IFREG;
1969 error = security_inode_create(dir, dentry, mode);
1970 if (error)
1971 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001973 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001974 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 return error;
1976}
1977
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001978int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001980 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 struct inode *inode = dentry->d_inode;
1982 int error;
1983
1984 if (!inode)
1985 return -ENOENT;
1986
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001987 switch (inode->i_mode & S_IFMT) {
1988 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001990 case S_IFDIR:
1991 if (acc_mode & MAY_WRITE)
1992 return -EISDIR;
1993 break;
1994 case S_IFBLK:
1995 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001996 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001998 /*FALLTHRU*/
1999 case S_IFIFO:
2000 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002002 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002003 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002004
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002005 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002006 if (error)
2007 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 /*
2010 * An append-only file must be opened in append mode for writing.
2011 */
2012 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002013 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002014 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002016 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 }
2018
2019 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05002020 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2021 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
2023 /*
2024 * Ensure there are no outstanding leases on the file.
2025 */
Al Virob65a9cf2009-12-16 06:27:40 -05002026 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002027}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Jeff Laytone1181ee2010-12-07 16:19:50 -05002029static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002030{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002031 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002032 struct inode *inode = path->dentry->d_inode;
2033 int error = get_write_access(inode);
2034 if (error)
2035 return error;
2036 /*
2037 * Refuse to truncate files with mandatory locks held on them.
2038 */
2039 error = locks_verify_locked(inode);
2040 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002041 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002042 if (!error) {
2043 error = do_truncate(path->dentry, 0,
2044 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002045 filp);
Al Viro7715b522009-12-16 03:54:00 -05002046 }
2047 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002048 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049}
2050
Dave Hansend57999e2008-02-15 14:37:27 -08002051/*
2052 * Be careful about ever adding any more callers of this
2053 * function. Its flags must be in the namei format, not
2054 * what get passed to sys_open().
2055 */
2056static int __open_namei_create(struct nameidata *nd, struct path *path,
Al Viro8737c932009-12-24 06:47:55 -05002057 int open_flag, int mode)
Dave Hansenaab520e2006-09-30 23:29:02 -07002058{
2059 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08002060 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07002061
2062 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002063 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002064 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2065 if (error)
2066 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07002067 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002068out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07002069 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08002070 dput(nd->path.dentry);
2071 nd->path.dentry = path->dentry;
Al Viro9b44f1b2011-03-09 00:17:27 -05002072 return error;
Dave Hansenaab520e2006-09-30 23:29:02 -07002073}
2074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075/*
Dave Hansend57999e2008-02-15 14:37:27 -08002076 * Note that while the flag value (low two bits) for sys_open means:
2077 * 00 - read-only
2078 * 01 - write-only
2079 * 10 - read-write
2080 * 11 - special
2081 * it is changed into
2082 * 00 - no permissions needed
2083 * 01 - read-permission
2084 * 10 - write-permission
2085 * 11 - read-write
2086 * for the internal routines (ie open_namei()/follow_link() etc)
2087 * This is more logical, and also allows the 00 "no perm needed"
2088 * to be used for symlinks (where the permissions are checked
2089 * later).
2090 *
2091*/
2092static inline int open_to_namei_flags(int flag)
2093{
2094 if ((flag+1) & O_ACCMODE)
2095 flag++;
2096 return flag;
2097}
2098
Al Viro7715b522009-12-16 03:54:00 -05002099static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08002100{
2101 /*
2102 * We'll never write to the fs underlying
2103 * a device file.
2104 */
2105 if (special_file(inode->i_mode))
2106 return 0;
2107 return (flag & O_TRUNC);
2108}
2109
Nick Piggin31e6b012011-01-07 17:49:52 +11002110/*
Al Virofe2d35f2011-03-05 22:58:25 -05002111 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002112 */
Al Virofb1cc552009-12-24 01:58:28 -05002113static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002114 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002115{
Al Viroa1e28032009-12-24 02:12:06 -05002116 struct dentry *dir = nd->path.dentry;
Al Viro0f9d1a12011-03-09 00:13:14 -05002117 int will_truncate;
Al Virofb1cc552009-12-24 01:58:28 -05002118 struct file *filp;
Al Virofe2d35f2011-03-05 22:58:25 -05002119 struct inode *inode;
Al Viro16c2cd72011-02-22 15:50:10 -05002120 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002121
Al Viroc3e380b2011-02-23 13:39:45 -05002122 nd->flags &= ~LOOKUP_PARENT;
2123 nd->flags |= op->intent;
2124
Al Viro1f36f772009-12-26 10:56:19 -05002125 switch (nd->last_type) {
2126 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002127 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002128 error = handle_dots(nd, nd->last_type);
2129 if (error)
2130 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002131 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002132 case LAST_ROOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002133 if (nd->flags & LOOKUP_RCU) {
2134 if (nameidata_drop_rcu_last(nd))
2135 return ERR_PTR(-ECHILD);
2136 }
Al Viro16c2cd72011-02-22 15:50:10 -05002137 error = handle_reval_path(nd);
2138 if (error)
2139 goto exit;
Al Virofe2d35f2011-03-05 22:58:25 -05002140 audit_inode(pathname, nd->path.dentry);
2141 if (op->open_flag & O_CREAT) {
2142 error = -EISDIR;
2143 goto exit;
2144 }
2145 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002146 case LAST_BIND:
Al Virofe2d35f2011-03-05 22:58:25 -05002147 /* can't be RCU mode here */
Al Viro16c2cd72011-02-22 15:50:10 -05002148 error = handle_reval_path(nd);
2149 if (error)
2150 goto exit;
Al Viro1f36f772009-12-26 10:56:19 -05002151 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002152 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002153 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002154
Al Virofe2d35f2011-03-05 22:58:25 -05002155 if (!(op->open_flag & O_CREAT)) {
2156 if (nd->last.name[nd->last.len])
2157 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2158 /* we _can_ be in RCU mode here */
2159 error = do_lookup(nd, &nd->last, path, &inode);
2160 if (error) {
2161 terminate_walk(nd);
2162 return ERR_PTR(error);
2163 }
2164 if (!inode) {
2165 path_to_nameidata(path, nd);
2166 terminate_walk(nd);
2167 return ERR_PTR(-ENOENT);
2168 }
2169 if (unlikely(inode->i_op->follow_link)) {
2170 /* We drop rcu-walk here */
2171 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2172 return ERR_PTR(-ECHILD);
2173 return NULL;
2174 }
2175 path_to_nameidata(path, nd);
2176 nd->inode = inode;
2177 /* sayonara */
2178 if (nd->flags & LOOKUP_RCU) {
2179 if (nameidata_drop_rcu_last(nd))
2180 return ERR_PTR(-ECHILD);
2181 }
2182
2183 error = -ENOTDIR;
2184 if (nd->flags & LOOKUP_DIRECTORY) {
2185 if (!inode->i_op->lookup)
2186 goto exit;
2187 }
2188 audit_inode(pathname, nd->path.dentry);
2189 goto ok;
2190 }
2191
2192 /* create side of things */
2193
2194 if (nd->flags & LOOKUP_RCU) {
2195 if (nameidata_drop_rcu_last(nd))
2196 return ERR_PTR(-ECHILD);
2197 }
2198
2199 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002200 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002201 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002202 if (nd->last.name[nd->last.len])
2203 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002204
Al Viroa1e28032009-12-24 02:12:06 -05002205 mutex_lock(&dir->d_inode->i_mutex);
2206
2207 path->dentry = lookup_hash(nd);
2208 path->mnt = nd->path.mnt;
2209
Al Virofb1cc552009-12-24 01:58:28 -05002210 error = PTR_ERR(path->dentry);
2211 if (IS_ERR(path->dentry)) {
2212 mutex_unlock(&dir->d_inode->i_mutex);
2213 goto exit;
2214 }
2215
2216 if (IS_ERR(nd->intent.open.file)) {
2217 error = PTR_ERR(nd->intent.open.file);
2218 goto exit_mutex_unlock;
2219 }
2220
2221 /* Negative dentry, just create the file */
2222 if (!path->dentry->d_inode) {
2223 /*
2224 * This write is needed to ensure that a
2225 * ro->rw transition does not occur between
2226 * the time when the file is created and when
2227 * a permanent write count is taken through
2228 * the 'struct file' in nameidata_to_filp().
2229 */
2230 error = mnt_want_write(nd->path.mnt);
2231 if (error)
2232 goto exit_mutex_unlock;
Al Viroc3e380b2011-02-23 13:39:45 -05002233 error = __open_namei_create(nd, path, op->open_flag, op->mode);
Al Virofb1cc552009-12-24 01:58:28 -05002234 if (error) {
2235 mnt_drop_write(nd->path.mnt);
2236 goto exit;
2237 }
Al Viro9b44f1b2011-03-09 00:17:27 -05002238 /* Don't check for write permission, don't truncate */
2239 error = may_open(&nd->path, 0, op->open_flag & ~O_TRUNC);
2240 if (error) {
2241 mnt_drop_write(nd->path.mnt);
2242 goto exit;
2243 }
Al Virofb1cc552009-12-24 01:58:28 -05002244 filp = nameidata_to_filp(nd);
2245 mnt_drop_write(nd->path.mnt);
Al Virod893f1b2010-10-29 03:30:42 -04002246 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002247 if (!IS_ERR(filp)) {
Al Viroc3e380b2011-02-23 13:39:45 -05002248 error = ima_file_check(filp, op->acc_mode);
Al Virofb1cc552009-12-24 01:58:28 -05002249 if (error) {
2250 fput(filp);
2251 filp = ERR_PTR(error);
2252 }
2253 }
2254 return filp;
2255 }
2256
2257 /*
2258 * It already exists.
2259 */
2260 mutex_unlock(&dir->d_inode->i_mutex);
2261 audit_inode(pathname, path->dentry);
2262
2263 error = -EEXIST;
Al Viroc3e380b2011-02-23 13:39:45 -05002264 if (op->open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002265 goto exit_dput;
2266
David Howells9875cf82011-01-14 18:45:21 +00002267 error = follow_managed(path, nd->flags);
2268 if (error < 0)
2269 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002270
2271 error = -ENOENT;
2272 if (!path->dentry->d_inode)
2273 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002274
2275 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002276 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002277
2278 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002279 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002280 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002281 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002282 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002283ok:
Al Viro0f9d1a12011-03-09 00:13:14 -05002284 will_truncate = open_will_truncate(op->open_flag, nd->path.dentry->d_inode);
2285 if (will_truncate) {
2286 error = mnt_want_write(nd->path.mnt);
2287 if (error)
2288 goto exit;
2289 }
2290 error = may_open(&nd->path, op->acc_mode, op->open_flag);
2291 if (error) {
2292 if (will_truncate)
2293 mnt_drop_write(nd->path.mnt);
2294 goto exit;
2295 }
2296 filp = nameidata_to_filp(nd);
2297 if (!IS_ERR(filp)) {
2298 error = ima_file_check(filp, op->acc_mode);
2299 if (error) {
2300 fput(filp);
2301 filp = ERR_PTR(error);
2302 }
2303 }
2304 if (!IS_ERR(filp)) {
2305 if (will_truncate) {
2306 error = handle_truncate(filp);
2307 if (error) {
2308 fput(filp);
2309 filp = ERR_PTR(error);
2310 }
2311 }
2312 }
2313 /*
2314 * It is now safe to drop the mnt write
2315 * because the filp has had a write taken
2316 * on its behalf.
2317 */
2318 if (will_truncate)
2319 mnt_drop_write(nd->path.mnt);
2320 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002321 return filp;
2322
2323exit_mutex_unlock:
2324 mutex_unlock(&dir->d_inode->i_mutex);
2325exit_dput:
2326 path_put_conditional(path, nd);
2327exit:
Al Virofb1cc552009-12-24 01:58:28 -05002328 path_put(&nd->path);
2329 return ERR_PTR(error);
2330}
2331
Al Viro13aab422011-02-23 17:54:08 -05002332static struct file *path_openat(int dfd, const char *pathname,
Al Viro47c805d2011-02-23 17:44:09 -05002333 const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334{
Al Virofe2d35f2011-03-05 22:58:25 -05002335 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002336 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08002337 struct nameidata nd;
Al Viro9850c052010-01-13 15:01:15 -05002338 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 int count = 0;
Al Viro13aab422011-02-23 17:54:08 -05002340 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002341
2342 filp = get_empty_filp();
2343 if (!filp)
2344 return ERR_PTR(-ENFILE);
2345
Al Viro47c805d2011-02-23 17:44:09 -05002346 filp->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11002347 nd.intent.open.file = filp;
Al Viro47c805d2011-02-23 17:44:09 -05002348 nd.intent.open.flags = open_to_namei_flags(op->open_flag);
2349 nd.intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002350
Al Virofe2d35f2011-03-05 22:58:25 -05002351 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, &nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002352 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002353 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002354
Al Virofe2d35f2011-03-05 22:58:25 -05002355 current->total_link_count = 0;
2356 error = link_path_walk(pathname, &nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002357 if (unlikely(error))
2358 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Al Viro47c805d2011-02-23 17:44:09 -05002360 filp = do_last(&nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002361 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002362 struct path link = path;
2363 struct inode *linki = link.dentry->d_inode;
Al Virodef4af32009-12-26 08:37:05 -05002364 void *cookie;
Al Viro806b6812009-12-26 07:16:40 -05002365 error = -ELOOP;
David Howellsdb372912011-01-14 18:45:53 +00002366 if (!(nd.flags & LOOKUP_FOLLOW))
Al Viro1f36f772009-12-26 10:56:19 -05002367 goto exit_dput;
2368 if (count++ == 32)
Al Viro806b6812009-12-26 07:16:40 -05002369 goto exit_dput;
2370 /*
2371 * This is subtle. Instead of calling do_follow_link() we do
2372 * the thing by hands. The reason is that this way we have zero
2373 * link_count and path_walk() (called from ->follow_link)
2374 * honoring LOOKUP_PARENT. After that we have the parent and
2375 * last component, i.e. we are in the same situation as after
2376 * the first path_walk(). Well, almost - if the last component
2377 * is normal we get its copy stored in nd->last.name and we will
2378 * have to putname() it when we are done. Procfs-like symlinks
2379 * just set LAST_BIND.
2380 */
2381 nd.flags |= LOOKUP_PARENT;
Al Viroc3e380b2011-02-23 13:39:45 -05002382 nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Nick Piggin7b9337a2011-01-14 08:42:43 +00002383 error = __do_follow_link(&link, &nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002384 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002385 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002386 else
Al Viro47c805d2011-02-23 17:44:09 -05002387 filp = do_last(&nd, &path, op, pathname);
Al Virof1afe9e2011-02-22 22:27:28 -05002388 if (!IS_ERR(cookie) && linki->i_op->put_link)
Nick Piggin7b9337a2011-01-14 08:42:43 +00002389 linki->i_op->put_link(link.dentry, &nd, cookie);
2390 path_put(&link);
Al Viro806b6812009-12-26 07:16:40 -05002391 }
Al Viro10fa8e62009-12-26 07:09:49 -05002392out:
Al Viro2a737872009-04-07 11:49:53 -04002393 if (nd.root.mnt)
2394 path_put(&nd.root);
Al Virofe2d35f2011-03-05 22:58:25 -05002395 if (base)
2396 fput(base);
Linus Torvalds2dab5972011-02-11 15:53:38 -08002397 release_open_intent(&nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002398 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Al Viro806b6812009-12-26 07:16:40 -05002400exit_dput:
2401 path_put_conditional(&path, &nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002402 path_put(&nd.path);
2403out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002404 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002405 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406}
2407
Al Viro13aab422011-02-23 17:54:08 -05002408struct file *do_filp_open(int dfd, const char *pathname,
2409 const struct open_flags *op, int flags)
2410{
2411 struct file *filp;
2412
2413 filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU);
2414 if (unlikely(filp == ERR_PTR(-ECHILD)))
2415 filp = path_openat(dfd, pathname, op, flags);
2416 if (unlikely(filp == ERR_PTR(-ESTALE)))
2417 filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL);
2418 return filp;
2419}
2420
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421/**
2422 * lookup_create - lookup a dentry, creating it if it doesn't exist
2423 * @nd: nameidata info
2424 * @is_dir: directory flag
2425 *
2426 * Simple function to lookup and return a dentry and create it
2427 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002428 *
Jan Blunck4ac91372008-02-14 19:34:32 -08002429 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 */
2431struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2432{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002433 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
Jan Blunck4ac91372008-02-14 19:34:32 -08002435 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002436 /*
2437 * Yucky last component or no last component at all?
2438 * (foo/., foo/.., /////)
2439 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 if (nd->last_type != LAST_NORM)
2441 goto fail;
2442 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04002443 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04002444 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002445
2446 /*
2447 * Do the final lookup.
2448 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08002449 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (IS_ERR(dentry))
2451 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002452
Al Viroe9baf6e2008-05-15 04:49:12 -04002453 if (dentry->d_inode)
2454 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002455 /*
2456 * Special case - lookup gave negative, but... we had foo/bar/
2457 * From the vfs_mknod() POV we just have a negative dentry -
2458 * all is fine. Let's be bastards - you had / on the end, you've
2459 * been asking for (non-existent) directory. -ENOENT for you.
2460 */
Al Viroe9baf6e2008-05-15 04:49:12 -04002461 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2462 dput(dentry);
2463 dentry = ERR_PTR(-ENOENT);
2464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002466eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002468 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469fail:
2470 return dentry;
2471}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07002472EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473
2474int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2475{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002476 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
2478 if (error)
2479 return error;
2480
2481 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2482 return -EPERM;
2483
Al Viroacfa4382008-12-04 10:06:33 -05002484 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 return -EPERM;
2486
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002487 error = devcgroup_inode_mknod(mode, dev);
2488 if (error)
2489 return error;
2490
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 error = security_inode_mknod(dir, dentry, mode, dev);
2492 if (error)
2493 return error;
2494
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002496 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002497 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 return error;
2499}
2500
Dave Hansen463c3192008-02-15 14:37:57 -08002501static int may_mknod(mode_t mode)
2502{
2503 switch (mode & S_IFMT) {
2504 case S_IFREG:
2505 case S_IFCHR:
2506 case S_IFBLK:
2507 case S_IFIFO:
2508 case S_IFSOCK:
2509 case 0: /* zero mode translates to S_IFREG */
2510 return 0;
2511 case S_IFDIR:
2512 return -EPERM;
2513 default:
2514 return -EINVAL;
2515 }
2516}
2517
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002518SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2519 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520{
Al Viro2ad94ae2008-07-21 09:32:51 -04002521 int error;
2522 char *tmp;
2523 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 struct nameidata nd;
2525
2526 if (S_ISDIR(mode))
2527 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Al Viro2ad94ae2008-07-21 09:32:51 -04002529 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002531 return error;
2532
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002534 if (IS_ERR(dentry)) {
2535 error = PTR_ERR(dentry);
2536 goto out_unlock;
2537 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002538 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002539 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002540 error = may_mknod(mode);
2541 if (error)
2542 goto out_dput;
2543 error = mnt_want_write(nd.path.mnt);
2544 if (error)
2545 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002546 error = security_path_mknod(&nd.path, dentry, mode, dev);
2547 if (error)
2548 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002549 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002551 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 break;
2553 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002554 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 new_decode_dev(dev));
2556 break;
2557 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002558 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002561out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002562 mnt_drop_write(nd.path.mnt);
2563out_dput:
2564 dput(dentry);
2565out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002566 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002567 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 putname(tmp);
2569
2570 return error;
2571}
2572
Heiko Carstens3480b252009-01-14 14:14:16 +01002573SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002574{
2575 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2576}
2577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2579{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002580 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
2582 if (error)
2583 return error;
2584
Al Viroacfa4382008-12-04 10:06:33 -05002585 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 return -EPERM;
2587
2588 mode &= (S_IRWXUGO|S_ISVTX);
2589 error = security_inode_mkdir(dir, dentry, mode);
2590 if (error)
2591 return error;
2592
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002594 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002595 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 return error;
2597}
2598
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002599SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600{
2601 int error = 0;
2602 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002603 struct dentry *dentry;
2604 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Al Viro2ad94ae2008-07-21 09:32:51 -04002606 error = user_path_parent(dfd, pathname, &nd, &tmp);
2607 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002608 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Dave Hansen6902d922006-09-30 23:29:01 -07002610 dentry = lookup_create(&nd, 1);
2611 error = PTR_ERR(dentry);
2612 if (IS_ERR(dentry))
2613 goto out_unlock;
2614
Jan Blunck4ac91372008-02-14 19:34:32 -08002615 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002616 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002617 error = mnt_want_write(nd.path.mnt);
2618 if (error)
2619 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002620 error = security_path_mkdir(&nd.path, dentry, mode);
2621 if (error)
2622 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002623 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002624out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002625 mnt_drop_write(nd.path.mnt);
2626out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002627 dput(dentry);
2628out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002629 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002630 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002631 putname(tmp);
2632out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 return error;
2634}
2635
Heiko Carstens3cdad422009-01-14 14:14:22 +01002636SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002637{
2638 return sys_mkdirat(AT_FDCWD, pathname, mode);
2639}
2640
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641/*
2642 * We try to drop the dentry early: we should have
2643 * a usage count of 2 if we're the only user of this
2644 * dentry, and if that is true (possibly after pruning
2645 * the dcache), then we drop the dentry now.
2646 *
2647 * A low-level filesystem can, if it choses, legally
2648 * do a
2649 *
2650 * if (!d_unhashed(dentry))
2651 * return -EBUSY;
2652 *
2653 * if it cannot handle the case of removing a directory
2654 * that is still in use by something else..
2655 */
2656void dentry_unhash(struct dentry *dentry)
2657{
2658 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002659 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002661 if (dentry->d_count == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 __d_drop(dentry);
2663 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664}
2665
2666int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2667{
2668 int error = may_delete(dir, dentry, 1);
2669
2670 if (error)
2671 return error;
2672
Al Viroacfa4382008-12-04 10:06:33 -05002673 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return -EPERM;
2675
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002676 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 dentry_unhash(dentry);
2678 if (d_mountpoint(dentry))
2679 error = -EBUSY;
2680 else {
2681 error = security_inode_rmdir(dir, dentry);
2682 if (!error) {
2683 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002684 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002686 dont_mount(dentry);
2687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 }
2689 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002690 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 d_delete(dentry);
2693 }
2694 dput(dentry);
2695
2696 return error;
2697}
2698
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002699static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700{
2701 int error = 0;
2702 char * name;
2703 struct dentry *dentry;
2704 struct nameidata nd;
2705
Al Viro2ad94ae2008-07-21 09:32:51 -04002706 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002708 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709
2710 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002711 case LAST_DOTDOT:
2712 error = -ENOTEMPTY;
2713 goto exit1;
2714 case LAST_DOT:
2715 error = -EINVAL;
2716 goto exit1;
2717 case LAST_ROOT:
2718 error = -EBUSY;
2719 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002721
2722 nd.flags &= ~LOOKUP_PARENT;
2723
Jan Blunck4ac91372008-02-14 19:34:32 -08002724 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002725 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002727 if (IS_ERR(dentry))
2728 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002729 error = mnt_want_write(nd.path.mnt);
2730 if (error)
2731 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002732 error = security_path_rmdir(&nd.path, dentry);
2733 if (error)
2734 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002735 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002736exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002737 mnt_drop_write(nd.path.mnt);
2738exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002739 dput(dentry);
2740exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002741 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002743 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 putname(name);
2745 return error;
2746}
2747
Heiko Carstens3cdad422009-01-14 14:14:22 +01002748SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002749{
2750 return do_rmdir(AT_FDCWD, pathname);
2751}
2752
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753int vfs_unlink(struct inode *dir, struct dentry *dentry)
2754{
2755 int error = may_delete(dir, dentry, 0);
2756
2757 if (error)
2758 return error;
2759
Al Viroacfa4382008-12-04 10:06:33 -05002760 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 return -EPERM;
2762
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002763 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 if (d_mountpoint(dentry))
2765 error = -EBUSY;
2766 else {
2767 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002768 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002770 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002771 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002774 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775
2776 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2777 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002778 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 }
Robert Love0eeca282005-07-12 17:06:03 -04002781
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 return error;
2783}
2784
2785/*
2786 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002787 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 * writeout happening, and we don't want to prevent access to the directory
2789 * while waiting on the I/O.
2790 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002791static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792{
Al Viro2ad94ae2008-07-21 09:32:51 -04002793 int error;
2794 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 struct dentry *dentry;
2796 struct nameidata nd;
2797 struct inode *inode = NULL;
2798
Al Viro2ad94ae2008-07-21 09:32:51 -04002799 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002801 return error;
2802
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 error = -EISDIR;
2804 if (nd.last_type != LAST_NORM)
2805 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002806
2807 nd.flags &= ~LOOKUP_PARENT;
2808
Jan Blunck4ac91372008-02-14 19:34:32 -08002809 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002810 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 error = PTR_ERR(dentry);
2812 if (!IS_ERR(dentry)) {
2813 /* Why not before? Because we want correct error value */
2814 if (nd.last.name[nd.last.len])
2815 goto slashes;
2816 inode = dentry->d_inode;
2817 if (inode)
Al Viro7de9c6e2010-10-23 11:11:40 -04002818 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002819 error = mnt_want_write(nd.path.mnt);
2820 if (error)
2821 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002822 error = security_path_unlink(&nd.path, dentry);
2823 if (error)
2824 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002825 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002826exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002827 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 exit2:
2829 dput(dentry);
2830 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002831 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 if (inode)
2833 iput(inode); /* truncate the inode here */
2834exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002835 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 putname(name);
2837 return error;
2838
2839slashes:
2840 error = !dentry->d_inode ? -ENOENT :
2841 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2842 goto exit2;
2843}
2844
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002845SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002846{
2847 if ((flag & ~AT_REMOVEDIR) != 0)
2848 return -EINVAL;
2849
2850 if (flag & AT_REMOVEDIR)
2851 return do_rmdir(dfd, pathname);
2852
2853 return do_unlinkat(dfd, pathname);
2854}
2855
Heiko Carstens3480b252009-01-14 14:14:16 +01002856SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002857{
2858 return do_unlinkat(AT_FDCWD, pathname);
2859}
2860
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002861int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002863 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
2865 if (error)
2866 return error;
2867
Al Viroacfa4382008-12-04 10:06:33 -05002868 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 return -EPERM;
2870
2871 error = security_inode_symlink(dir, dentry, oldname);
2872 if (error)
2873 return error;
2874
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002876 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002877 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 return error;
2879}
2880
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002881SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2882 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883{
Al Viro2ad94ae2008-07-21 09:32:51 -04002884 int error;
2885 char *from;
2886 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002887 struct dentry *dentry;
2888 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
2890 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002891 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002893
2894 error = user_path_parent(newdfd, newname, &nd, &to);
2895 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002896 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897
Dave Hansen6902d922006-09-30 23:29:01 -07002898 dentry = lookup_create(&nd, 0);
2899 error = PTR_ERR(dentry);
2900 if (IS_ERR(dentry))
2901 goto out_unlock;
2902
Dave Hansen75c3f292008-02-15 14:37:45 -08002903 error = mnt_want_write(nd.path.mnt);
2904 if (error)
2905 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002906 error = security_path_symlink(&nd.path, dentry, from);
2907 if (error)
2908 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002909 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002910out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002911 mnt_drop_write(nd.path.mnt);
2912out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002913 dput(dentry);
2914out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002915 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002916 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002917 putname(to);
2918out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 putname(from);
2920 return error;
2921}
2922
Heiko Carstens3480b252009-01-14 14:14:16 +01002923SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002924{
2925 return sys_symlinkat(oldname, AT_FDCWD, newname);
2926}
2927
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2929{
2930 struct inode *inode = old_dentry->d_inode;
2931 int error;
2932
2933 if (!inode)
2934 return -ENOENT;
2935
Miklos Szeredia95164d2008-07-30 15:08:48 +02002936 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 if (error)
2938 return error;
2939
2940 if (dir->i_sb != inode->i_sb)
2941 return -EXDEV;
2942
2943 /*
2944 * A link to an append-only or immutable file cannot be created.
2945 */
2946 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2947 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002948 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002950 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 return -EPERM;
2952
2953 error = security_inode_link(old_dentry, dir, new_dentry);
2954 if (error)
2955 return error;
2956
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002957 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002959 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002960 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002961 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 return error;
2963}
2964
2965/*
2966 * Hardlinks are often used in delicate situations. We avoid
2967 * security-related surprises by not following symlinks on the
2968 * newname. --KAB
2969 *
2970 * We don't follow them on the oldname either to be compatible
2971 * with linux 2.0, and to avoid hard-linking to directories
2972 * and other special files. --ADM
2973 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002974SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2975 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976{
2977 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002978 struct nameidata nd;
2979 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002981 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002983 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002984 return -EINVAL;
2985
Al Viro2d8f3032008-07-22 09:59:21 -04002986 error = user_path_at(olddfd, oldname,
2987 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2988 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002990 return error;
2991
2992 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 if (error)
2994 goto out;
2995 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002996 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 goto out_release;
2998 new_dentry = lookup_create(&nd, 0);
2999 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003000 if (IS_ERR(new_dentry))
3001 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08003002 error = mnt_want_write(nd.path.mnt);
3003 if (error)
3004 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003005 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3006 if (error)
3007 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04003008 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003009out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08003010 mnt_drop_write(nd.path.mnt);
3011out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003012 dput(new_dentry);
3013out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08003014 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08003016 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003017 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018out:
Al Viro2d8f3032008-07-22 09:59:21 -04003019 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020
3021 return error;
3022}
3023
Heiko Carstens3480b252009-01-14 14:14:16 +01003024SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003025{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003026 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003027}
3028
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029/*
3030 * The worst of all namespace operations - renaming directory. "Perverted"
3031 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3032 * Problems:
3033 * a) we can get into loop creation. Check is done in is_subdir().
3034 * b) race potential - two innocent renames can create a loop together.
3035 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003036 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 * story.
3038 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003039 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 * whether the target exists). Solution: try to be smart with locking
3041 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003042 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 * move will be locked. Thus we can rank directories by the tree
3044 * (ancestors first) and rank all non-directories after them.
3045 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003046 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 * HOWEVER, it relies on the assumption that any object with ->lookup()
3048 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3049 * we'd better make sure that there's no link(2) for them.
3050 * d) some filesystems don't support opened-but-unlinked directories,
3051 * either because of layout or because they are not ready to deal with
3052 * all cases correctly. The latter will be fixed (taking this sort of
3053 * stuff into VFS), but the former is not going away. Solution: the same
3054 * trick as in rmdir().
3055 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003056 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003058 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 * locking].
3060 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003061static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3062 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063{
3064 int error = 0;
3065 struct inode *target;
3066
3067 /*
3068 * If we are going to change the parent - check write permissions,
3069 * we'll need to flip '..'.
3070 */
3071 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003072 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 if (error)
3074 return error;
3075 }
3076
3077 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3078 if (error)
3079 return error;
3080
3081 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04003082 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003083 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3085 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04003086 else {
3087 if (target)
3088 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04003092 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04003094 dont_mount(new_dentry);
3095 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003096 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 if (d_unhashed(new_dentry))
3098 d_rehash(new_dentry);
3099 dput(new_dentry);
3100 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003101 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003102 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3103 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 return error;
3105}
3106
Adrian Bunk75c96f82005-05-05 16:16:09 -07003107static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3108 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109{
3110 struct inode *target;
3111 int error;
3112
3113 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3114 if (error)
3115 return error;
3116
3117 dget(new_dentry);
3118 target = new_dentry->d_inode;
3119 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003120 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3122 error = -EBUSY;
3123 else
3124 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3125 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05003126 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04003127 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07003128 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 }
3131 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003132 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 dput(new_dentry);
3134 return error;
3135}
3136
3137int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3138 struct inode *new_dir, struct dentry *new_dentry)
3139{
3140 int error;
3141 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003142 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143
3144 if (old_dentry->d_inode == new_dentry->d_inode)
3145 return 0;
3146
3147 error = may_delete(old_dir, old_dentry, is_dir);
3148 if (error)
3149 return error;
3150
3151 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003152 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 else
3154 error = may_delete(new_dir, new_dentry, is_dir);
3155 if (error)
3156 return error;
3157
Al Viroacfa4382008-12-04 10:06:33 -05003158 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 return -EPERM;
3160
Robert Love0eeca282005-07-12 17:06:03 -04003161 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3162
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 if (is_dir)
3164 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3165 else
3166 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003167 if (!error)
3168 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003169 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003170 fsnotify_oldname_free(old_name);
3171
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 return error;
3173}
3174
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003175SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3176 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177{
Al Viro2ad94ae2008-07-21 09:32:51 -04003178 struct dentry *old_dir, *new_dir;
3179 struct dentry *old_dentry, *new_dentry;
3180 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003182 char *from;
3183 char *to;
3184 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185
Al Viro2ad94ae2008-07-21 09:32:51 -04003186 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 if (error)
3188 goto exit;
3189
Al Viro2ad94ae2008-07-21 09:32:51 -04003190 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 if (error)
3192 goto exit1;
3193
3194 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003195 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 goto exit2;
3197
Jan Blunck4ac91372008-02-14 19:34:32 -08003198 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 error = -EBUSY;
3200 if (oldnd.last_type != LAST_NORM)
3201 goto exit2;
3202
Jan Blunck4ac91372008-02-14 19:34:32 -08003203 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 if (newnd.last_type != LAST_NORM)
3205 goto exit2;
3206
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003207 oldnd.flags &= ~LOOKUP_PARENT;
3208 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003209 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003210
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 trap = lock_rename(new_dir, old_dir);
3212
Christoph Hellwig49705b72005-11-08 21:35:06 -08003213 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 error = PTR_ERR(old_dentry);
3215 if (IS_ERR(old_dentry))
3216 goto exit3;
3217 /* source must exist */
3218 error = -ENOENT;
3219 if (!old_dentry->d_inode)
3220 goto exit4;
3221 /* unless the source is a directory trailing slashes give -ENOTDIR */
3222 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3223 error = -ENOTDIR;
3224 if (oldnd.last.name[oldnd.last.len])
3225 goto exit4;
3226 if (newnd.last.name[newnd.last.len])
3227 goto exit4;
3228 }
3229 /* source should not be ancestor of target */
3230 error = -EINVAL;
3231 if (old_dentry == trap)
3232 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003233 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 error = PTR_ERR(new_dentry);
3235 if (IS_ERR(new_dentry))
3236 goto exit4;
3237 /* target should not be an ancestor of source */
3238 error = -ENOTEMPTY;
3239 if (new_dentry == trap)
3240 goto exit5;
3241
Dave Hansen9079b1e2008-02-15 14:37:49 -08003242 error = mnt_want_write(oldnd.path.mnt);
3243 if (error)
3244 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003245 error = security_path_rename(&oldnd.path, old_dentry,
3246 &newnd.path, new_dentry);
3247 if (error)
3248 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 error = vfs_rename(old_dir->d_inode, old_dentry,
3250 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003251exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003252 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253exit5:
3254 dput(new_dentry);
3255exit4:
3256 dput(old_dentry);
3257exit3:
3258 unlock_rename(new_dir, old_dir);
3259exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003260 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003261 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003263 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003265exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 return error;
3267}
3268
Heiko Carstensa26eab22009-01-14 14:14:17 +01003269SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003270{
3271 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3272}
3273
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3275{
3276 int len;
3277
3278 len = PTR_ERR(link);
3279 if (IS_ERR(link))
3280 goto out;
3281
3282 len = strlen(link);
3283 if (len > (unsigned) buflen)
3284 len = buflen;
3285 if (copy_to_user(buffer, link, len))
3286 len = -EFAULT;
3287out:
3288 return len;
3289}
3290
3291/*
3292 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3293 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3294 * using) it for any given inode is up to filesystem.
3295 */
3296int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3297{
3298 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003299 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003300 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003301
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003303 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003304 if (IS_ERR(cookie))
3305 return PTR_ERR(cookie);
3306
3307 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3308 if (dentry->d_inode->i_op->put_link)
3309 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3310 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311}
3312
3313int vfs_follow_link(struct nameidata *nd, const char *link)
3314{
3315 return __vfs_follow_link(nd, link);
3316}
3317
3318/* get the link contents into pagecache */
3319static char *page_getlink(struct dentry * dentry, struct page **ppage)
3320{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003321 char *kaddr;
3322 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003324 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003326 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003328 kaddr = kmap(page);
3329 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3330 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331}
3332
3333int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3334{
3335 struct page *page = NULL;
3336 char *s = page_getlink(dentry, &page);
3337 int res = vfs_readlink(dentry,buffer,buflen,s);
3338 if (page) {
3339 kunmap(page);
3340 page_cache_release(page);
3341 }
3342 return res;
3343}
3344
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003345void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003347 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003349 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350}
3351
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003352void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003354 struct page *page = cookie;
3355
3356 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 kunmap(page);
3358 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 }
3360}
3361
Nick Piggin54566b22009-01-04 12:00:53 -08003362/*
3363 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3364 */
3365int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366{
3367 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003368 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003369 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003370 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003372 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3373 if (nofs)
3374 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
NeilBrown7e53cac2006-03-25 03:07:57 -08003376retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003377 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003378 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003380 goto fail;
3381
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 kaddr = kmap_atomic(page, KM_USER0);
3383 memcpy(kaddr, symname, len-1);
3384 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003385
3386 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3387 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 if (err < 0)
3389 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003390 if (err < len-1)
3391 goto retry;
3392
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 mark_inode_dirty(inode);
3394 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395fail:
3396 return err;
3397}
3398
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003399int page_symlink(struct inode *inode, const char *symname, int len)
3400{
3401 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003402 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003403}
3404
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003405const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 .readlink = generic_readlink,
3407 .follow_link = page_follow_link_light,
3408 .put_link = page_put_link,
3409};
3410
Al Viro2d8f3032008-07-22 09:59:21 -04003411EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003412EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413EXPORT_SYMBOL(follow_down);
3414EXPORT_SYMBOL(follow_up);
3415EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3416EXPORT_SYMBOL(getname);
3417EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418EXPORT_SYMBOL(lookup_one_len);
3419EXPORT_SYMBOL(page_follow_link_light);
3420EXPORT_SYMBOL(page_put_link);
3421EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003422EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423EXPORT_SYMBOL(page_symlink);
3424EXPORT_SYMBOL(page_symlink_inode_operations);
Al Viroc9c6cac2011-02-16 15:15:47 -05003425EXPORT_SYMBOL(kern_path_parent);
Al Virod1811462008-08-02 00:49:18 -04003426EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003427EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003428EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08003429EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430EXPORT_SYMBOL(unlock_rename);
3431EXPORT_SYMBOL(vfs_create);
3432EXPORT_SYMBOL(vfs_follow_link);
3433EXPORT_SYMBOL(vfs_link);
3434EXPORT_SYMBOL(vfs_mkdir);
3435EXPORT_SYMBOL(vfs_mknod);
3436EXPORT_SYMBOL(generic_permission);
3437EXPORT_SYMBOL(vfs_readlink);
3438EXPORT_SYMBOL(vfs_rename);
3439EXPORT_SYMBOL(vfs_rmdir);
3440EXPORT_SYMBOL(vfs_symlink);
3441EXPORT_SYMBOL(vfs_unlink);
3442EXPORT_SYMBOL(dentry_unhash);
3443EXPORT_SYMBOL(generic_readlink);