blob: 14ab8d3f2f0c8f7fc3e829ed26404e53a2420028 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
Eric Parise81e3f42009-12-04 15:47:36 -050037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030073 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800117static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
Al Virof52e0c12011-03-14 18:56:51 -0400139static char *getname_flags(const char __user * filename, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
Al Virof52e0c12011-03-14 18:56:51 -0400150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
Al Virof52e0c12011-03-14 18:56:51 -0400160char *getname(const char __user * filename)
161{
162 return getname_flags(filename, 0);
163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165#ifdef CONFIG_AUDITSYSCALL
166void putname(const char *name)
167{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400168 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 audit_putname(name);
170 else
171 __putname(name);
172}
173EXPORT_SYMBOL(putname);
174#endif
175
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700176/*
177 * This does basic POSIX ACL permission checking
178 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100179static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700181{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700182 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700183
184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
185
Serge E. Hallyne795b712011-03-23 16:43:25 -0700186 if (current_user_ns() != inode_userns(inode))
187 goto other_perms;
188
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700189 if (current_fsuid() == inode->i_uid)
190 mode >>= 6;
191 else {
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100193 int error = check_acl(inode, mask, flags);
194 if (error != -EAGAIN)
195 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
Serge E. Hallyne795b712011-03-23 16:43:25 -0700202other_perms:
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700203 /*
204 * If the DACs are ok we don't need any capability check.
205 */
206 if ((mask & ~mode) == 0)
207 return 0;
208 return -EACCES;
209}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100212 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
Randy Dunlap39191622011-01-08 19:36:21 -0800216 * @flags: IPERM_FLAG_ flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 *
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100221 * are used for other things.
222 *
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100227int generic_permission(struct inode *inode, int mask, unsigned int flags,
228 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700230 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700233 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100235 ret = acl_permission_check(inode, mask, flags, check_acl);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700236 if (ret != -EACCES)
237 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 /*
240 * Read/write DACs are always overridable.
Al Viro8e833fd2011-06-19 01:56:53 -0400241 * Executable DACs are overridable for all directories and
242 * for non-directories that have least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200244 if (!(mask & MAY_EXEC) || execute_ok(inode))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700245 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 return 0;
247
248 /*
249 * Searching includes executable on directories, else just read.
250 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600251 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
Serge E. Hallyne795b712011-03-23 16:43:25 -0700253 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 return 0;
255
256 return -EACCES;
257}
258
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200259/**
260 * inode_permission - check for access rights to a given inode
261 * @inode: inode to check permission on
262 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 *
264 * Used to check for read/write/execute permissions on an inode.
265 * We use "fsuid" for this, letting us set arbitrary permissions
266 * for filesystem access without changing the "normal" uids which
267 * are used for other things.
268 */
Al Virof419a2e2008-07-22 00:07:17 -0400269int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Al Viroe6305c42008-07-15 21:03:57 -0400271 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
273 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700274 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276 /*
277 * Nobody gets write access to a read-only fs.
278 */
279 if (IS_RDONLY(inode) &&
280 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
281 return -EROFS;
282
283 /*
284 * Nobody gets write access to an immutable file.
285 */
286 if (IS_IMMUTABLE(inode))
287 return -EACCES;
288 }
289
Al Viroacfa4382008-12-04 10:06:33 -0500290 if (inode->i_op->permission)
Nick Pigginb74c79e2011-01-07 17:49:58 +1100291 retval = inode->i_op->permission(inode, mask, 0);
Miklos Szeredif696a362008-07-31 13:41:58 +0200292 else
Nick Pigginb74c79e2011-01-07 17:49:58 +1100293 retval = generic_permission(inode, mask, 0,
294 inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (retval)
297 return retval;
298
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700299 retval = devcgroup_inode_permission(inode, mask);
300 if (retval)
301 return retval;
302
Eric Parisd09ca732010-07-23 11:43:57 -0400303 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800306/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800307 * file_permission - check for additional access rights to a given file
308 * @file: file to check access rights for
309 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
310 *
311 * Used to check for read/write/execute permissions on an already opened
312 * file.
313 *
314 * Note:
315 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200316 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800317 */
318int file_permission(struct file *file, int mask)
319{
Al Virof419a2e2008-07-22 00:07:17 -0400320 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800321}
322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323/*
324 * get_write_access() gets write permission for a file.
325 * put_write_access() releases this write permission.
326 * This is used for regular files.
327 * We cannot support write (and maybe mmap read-write shared) accesses and
328 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
329 * can have the following values:
330 * 0: no writers, no VM_DENYWRITE mappings
331 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
332 * > 0: (i_writecount) users are writing to the file.
333 *
334 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
335 * except for the cases where we don't hold i_writecount yet. Then we need to
336 * use {get,deny}_write_access() - these functions check the sign and refuse
337 * to do the change if sign is wrong. Exclusion between them is provided by
338 * the inode->i_lock spinlock.
339 */
340
341int get_write_access(struct inode * inode)
342{
343 spin_lock(&inode->i_lock);
344 if (atomic_read(&inode->i_writecount) < 0) {
345 spin_unlock(&inode->i_lock);
346 return -ETXTBSY;
347 }
348 atomic_inc(&inode->i_writecount);
349 spin_unlock(&inode->i_lock);
350
351 return 0;
352}
353
354int deny_write_access(struct file * file)
355{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800356 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 spin_lock(&inode->i_lock);
359 if (atomic_read(&inode->i_writecount) > 0) {
360 spin_unlock(&inode->i_lock);
361 return -ETXTBSY;
362 }
363 atomic_dec(&inode->i_writecount);
364 spin_unlock(&inode->i_lock);
365
366 return 0;
367}
368
Jan Blunck1d957f92008-02-14 19:34:35 -0800369/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800370 * path_get - get a reference to a path
371 * @path: path to get the reference to
372 *
373 * Given a path increment the reference count to the dentry and the vfsmount.
374 */
375void path_get(struct path *path)
376{
377 mntget(path->mnt);
378 dget(path->dentry);
379}
380EXPORT_SYMBOL(path_get);
381
382/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800383 * path_put - put a reference to a path
384 * @path: path to put the reference to
385 *
386 * Given a path decrement the reference count to the dentry and the vfsmount.
387 */
388void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Jan Blunck1d957f92008-02-14 19:34:35 -0800390 dput(path->dentry);
391 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392}
Jan Blunck1d957f92008-02-14 19:34:35 -0800393EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Al Viro19660af2011-03-25 10:32:48 -0400395/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100396 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400397 * Documentation/filesystems/path-lookup.txt). In situations when we can't
398 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
399 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
400 * mode. Refcounts are grabbed at the last known good point before rcu-walk
401 * got stuck, so ref-walk may continue from there. If this is not successful
402 * (eg. a seqcount has changed), then failure is returned and it's up to caller
403 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100404 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100405
406/**
Al Viro19660af2011-03-25 10:32:48 -0400407 * unlazy_walk - try to switch to ref-walk mode.
408 * @nd: nameidata pathwalk data
409 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800410 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100411 *
Al Viro19660af2011-03-25 10:32:48 -0400412 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
413 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
414 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100415 */
Al Viro19660af2011-03-25 10:32:48 -0400416static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100417{
418 struct fs_struct *fs = current->fs;
419 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500420 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100421
422 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500423 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
424 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100425 spin_lock(&fs->lock);
426 if (nd->root.mnt != fs->root.mnt ||
427 nd->root.dentry != fs->root.dentry)
428 goto err_root;
429 }
430 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400431 if (!dentry) {
432 if (!__d_rcu_to_refcount(parent, nd->seq))
433 goto err_parent;
434 BUG_ON(nd->inode != parent->d_inode);
435 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400436 if (dentry->d_parent != parent)
437 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400438 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
439 if (!__d_rcu_to_refcount(dentry, nd->seq))
440 goto err_child;
441 /*
442 * If the sequence check on the child dentry passed, then
443 * the child has not been removed from its parent. This
444 * means the parent dentry must be valid and able to take
445 * a reference at this point.
446 */
447 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
448 BUG_ON(!parent->d_count);
449 parent->d_count++;
450 spin_unlock(&dentry->d_lock);
451 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100452 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500453 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100454 path_get(&nd->root);
455 spin_unlock(&fs->lock);
456 }
457 mntget(nd->path.mnt);
458
459 rcu_read_unlock();
460 br_read_unlock(vfsmount_lock);
461 nd->flags &= ~LOOKUP_RCU;
462 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400463
464err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100465 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400466err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100467 spin_unlock(&parent->d_lock);
468err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500469 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100470 spin_unlock(&fs->lock);
471 return -ECHILD;
472}
473
Nick Piggin31e6b012011-01-07 17:49:52 +1100474/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700475 * release_open_intent - free up open intent resources
476 * @nd: pointer to nameidata
477 */
478void release_open_intent(struct nameidata *nd)
479{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800480 struct file *file = nd->intent.open.file;
481
482 if (file && !IS_ERR(file)) {
483 if (file->f_path.dentry == NULL)
484 put_filp(file);
485 else
486 fput(file);
487 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700488}
489
Al Virof60aef72011-02-15 01:35:28 -0500490static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100491{
Al Virof60aef72011-02-15 01:35:28 -0500492 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100493}
494
Al Virof5e1c1c2011-02-15 01:32:55 -0500495static struct dentry *
Ian Kentbcdc5e02006-09-27 01:50:44 -0700496do_revalidate(struct dentry *dentry, struct nameidata *nd)
497{
Al Virof5e1c1c2011-02-15 01:32:55 -0500498 int status = d_revalidate(dentry, nd);
Ian Kentbcdc5e02006-09-27 01:50:44 -0700499 if (unlikely(status <= 0)) {
500 /*
501 * The dentry failed validation.
502 * If d_revalidate returned 0 attempt to invalidate
503 * the dentry otherwise d_revalidate is asking us
504 * to return a fail status.
505 */
Nick Piggin34286d62011-01-07 17:49:57 +1100506 if (status < 0) {
Al Virof5e1c1c2011-02-15 01:32:55 -0500507 dput(dentry);
Nick Piggin34286d62011-01-07 17:49:57 +1100508 dentry = ERR_PTR(status);
Al Virof5e1c1c2011-02-15 01:32:55 -0500509 } else if (!d_invalidate(dentry)) {
510 dput(dentry);
511 dentry = NULL;
Ian Kentbcdc5e02006-09-27 01:50:44 -0700512 }
513 }
514 return dentry;
515}
516
Al Viro9f1fafe2011-03-25 11:00:12 -0400517/**
518 * complete_walk - successful completion of path walk
519 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500520 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400521 * If we had been in RCU mode, drop out of it and legitimize nd->path.
522 * Revalidate the final result, unless we'd already done that during
523 * the path walk or the filesystem doesn't ask for it. Return 0 on
524 * success, -error on failure. In case of failure caller does not
525 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500526 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400527static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500528{
Al Viro16c2cd72011-02-22 15:50:10 -0500529 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500530 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500531
Al Viro9f1fafe2011-03-25 11:00:12 -0400532 if (nd->flags & LOOKUP_RCU) {
533 nd->flags &= ~LOOKUP_RCU;
534 if (!(nd->flags & LOOKUP_ROOT))
535 nd->root.mnt = NULL;
536 spin_lock(&dentry->d_lock);
537 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
538 spin_unlock(&dentry->d_lock);
539 rcu_read_unlock();
540 br_read_unlock(vfsmount_lock);
541 return -ECHILD;
542 }
543 BUG_ON(nd->inode != dentry->d_inode);
544 spin_unlock(&dentry->d_lock);
545 mntget(nd->path.mnt);
546 rcu_read_unlock();
547 br_read_unlock(vfsmount_lock);
548 }
549
Al Viro16c2cd72011-02-22 15:50:10 -0500550 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500551 return 0;
552
Al Viro16c2cd72011-02-22 15:50:10 -0500553 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
554 return 0;
555
556 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
557 return 0;
558
559 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100560 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500561 if (status > 0)
562 return 0;
563
Al Viro16c2cd72011-02-22 15:50:10 -0500564 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500565 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500566
Al Viro9f1fafe2011-03-25 11:00:12 -0400567 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500568 return status;
569}
570
571/*
Al Virob75b5082009-12-16 01:01:38 -0500572 * Short-cut version of permission(), for calling on directories
573 * during pathname resolution. Combines parts of permission()
574 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 *
576 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500577 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 * complete permission check.
579 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100580static inline int exec_permission(struct inode *inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700582 int ret;
Serge E. Hallyne795b712011-03-23 16:43:25 -0700583 struct user_namespace *ns = inode_userns(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700585 if (inode->i_op->permission) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100586 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
587 } else {
588 ret = acl_permission_check(inode, MAY_EXEC, flags,
589 inode->i_op->check_acl);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700590 }
Nick Pigginb74c79e2011-01-07 17:49:58 +1100591 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 goto ok;
Nick Pigginb74c79e2011-01-07 17:49:58 +1100593 if (ret == -ECHILD)
Nick Piggin31e6b012011-01-07 17:49:52 +1100594 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Serge E. Hallyne795b712011-03-23 16:43:25 -0700596 if (ns_capable(ns, CAP_DAC_OVERRIDE) ||
597 ns_capable(ns, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 goto ok;
599
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700600 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601ok:
Nick Pigginb74c79e2011-01-07 17:49:58 +1100602 return security_inode_exec_permission(inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
Al Viro2a737872009-04-07 11:49:53 -0400605static __always_inline void set_root(struct nameidata *nd)
606{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200607 if (!nd->root.mnt)
608 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400609}
610
Al Viro6de88d72009-08-09 01:41:57 +0400611static int link_path_walk(const char *, struct nameidata *);
612
Nick Piggin31e6b012011-01-07 17:49:52 +1100613static __always_inline void set_root_rcu(struct nameidata *nd)
614{
615 if (!nd->root.mnt) {
616 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100617 unsigned seq;
618
619 do {
620 seq = read_seqcount_begin(&fs->seq);
621 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700622 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100623 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100624 }
625}
626
Arjan van de Venf1662352006-01-14 13:21:31 -0800627static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Nick Piggin31e6b012011-01-07 17:49:52 +1100629 int ret;
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 if (IS_ERR(link))
632 goto fail;
633
634 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400635 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800636 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400637 nd->path = nd->root;
638 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500639 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100641 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100642
Nick Piggin31e6b012011-01-07 17:49:52 +1100643 ret = link_path_walk(link, nd);
644 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800646 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 return PTR_ERR(link);
648}
649
Jan Blunck1d957f92008-02-14 19:34:35 -0800650static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700651{
652 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800653 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700654 mntput(path->mnt);
655}
656
Nick Piggin7b9337a2011-01-14 08:42:43 +0000657static inline void path_to_nameidata(const struct path *path,
658 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700659{
Nick Piggin31e6b012011-01-07 17:49:52 +1100660 if (!(nd->flags & LOOKUP_RCU)) {
661 dput(nd->path.dentry);
662 if (nd->path.mnt != path->mnt)
663 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800664 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100665 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800666 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700667}
668
Al Viro574197e2011-03-14 22:20:34 -0400669static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
670{
671 struct inode *inode = link->dentry->d_inode;
672 if (!IS_ERR(cookie) && inode->i_op->put_link)
673 inode->i_op->put_link(link->dentry, nd, cookie);
674 path_put(link);
675}
676
Al Virodef4af32009-12-26 08:37:05 -0500677static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400678follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800679{
680 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000681 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800682
Al Viro844a3912011-02-15 00:38:26 -0500683 BUG_ON(nd->flags & LOOKUP_RCU);
684
Al Viro0e794582011-03-16 02:45:02 -0400685 if (link->mnt == nd->path.mnt)
686 mntget(link->mnt);
687
Al Viro574197e2011-03-14 22:20:34 -0400688 if (unlikely(current->total_link_count >= 40)) {
689 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
Al Viro574197e2011-03-14 22:20:34 -0400690 path_put(&nd->path);
691 return -ELOOP;
692 }
693 cond_resched();
694 current->total_link_count++;
695
Nick Piggin7b9337a2011-01-14 08:42:43 +0000696 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800697 nd_set_link(nd, NULL);
698
Al Viro36f3b4f2011-02-22 21:24:38 -0500699 error = security_inode_follow_link(link->dentry, nd);
700 if (error) {
701 *p = ERR_PTR(error); /* no ->put_link(), please */
702 path_put(&nd->path);
703 return error;
704 }
705
Al Viro86acdca12009-12-22 23:45:11 -0500706 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500707 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
708 error = PTR_ERR(*p);
709 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800710 char *s = nd_get_link(nd);
711 error = 0;
712 if (s)
713 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400714 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500715 nd->flags |= LOOKUP_JUMPED;
Al Virob21041d2011-03-14 20:01:51 -0400716 nd->inode = nd->path.dentry->d_inode;
717 if (nd->inode->i_op->follow_link) {
Al Virobcda7652011-03-13 16:42:14 -0400718 /* stepped on a _really_ weird one */
719 path_put(&nd->path);
720 error = -ELOOP;
721 }
722 }
Ian Kent051d3812006-03-27 01:14:53 -0800723 }
Ian Kent051d3812006-03-27 01:14:53 -0800724 return error;
725}
726
Nick Piggin31e6b012011-01-07 17:49:52 +1100727static int follow_up_rcu(struct path *path)
728{
729 struct vfsmount *parent;
730 struct dentry *mountpoint;
731
732 parent = path->mnt->mnt_parent;
733 if (parent == path->mnt)
734 return 0;
735 mountpoint = path->mnt->mnt_mountpoint;
736 path->dentry = mountpoint;
737 path->mnt = parent;
738 return 1;
739}
740
Al Virobab77eb2009-04-18 03:26:48 -0400741int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
743 struct vfsmount *parent;
744 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000745
746 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400747 parent = path->mnt->mnt_parent;
748 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000749 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return 0;
751 }
752 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400753 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000754 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400755 dput(path->dentry);
756 path->dentry = mountpoint;
757 mntput(path->mnt);
758 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return 1;
760}
761
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100762/*
David Howells9875cf82011-01-14 18:45:21 +0000763 * Perform an automount
764 * - return -EISDIR to tell follow_managed() to stop and return the path we
765 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 */
David Howells9875cf82011-01-14 18:45:21 +0000767static int follow_automount(struct path *path, unsigned flags,
768 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100769{
David Howells9875cf82011-01-14 18:45:21 +0000770 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000771 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100772
David Howells9875cf82011-01-14 18:45:21 +0000773 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
774 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700775
David Howells6f45b652011-01-14 18:45:31 +0000776 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
777 * and this is the terminal part of the path.
778 */
779 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
780 return -EISDIR; /* we actually want to stop here */
781
David Howells9875cf82011-01-14 18:45:21 +0000782 /* We want to mount if someone is trying to open/create a file of any
783 * type under the mountpoint, wants to traverse through the mountpoint
784 * or wants to open the mounted directory.
785 *
786 * We don't want to mount if someone's just doing a stat and they've
787 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
788 * appended a '/' to the name.
789 */
790 if (!(flags & LOOKUP_FOLLOW) &&
791 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
792 LOOKUP_OPEN | LOOKUP_CREATE)))
793 return -EISDIR;
794
795 current->total_link_count++;
796 if (current->total_link_count >= 40)
797 return -ELOOP;
798
799 mnt = path->dentry->d_op->d_automount(path);
800 if (IS_ERR(mnt)) {
801 /*
802 * The filesystem is allowed to return -EISDIR here to indicate
803 * it doesn't want to automount. For instance, autofs would do
804 * this so that its userspace daemon can mount on this dentry.
805 *
806 * However, we can only permit this if it's a terminal point in
807 * the path being looked up; if it wasn't then the remainder of
808 * the path is inaccessible and we should say so.
809 */
810 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
811 return -EREMOTE;
812 return PTR_ERR(mnt);
813 }
David Howellsea5b7782011-01-14 19:10:03 +0000814
David Howells9875cf82011-01-14 18:45:21 +0000815 if (!mnt) /* mount collision */
816 return 0;
817
Al Viro8aef1882011-06-16 15:10:06 +0100818 if (!*need_mntput) {
819 /* lock_mount() may release path->mnt on error */
820 mntget(path->mnt);
821 *need_mntput = true;
822 }
Al Viro19a167a2011-01-17 01:35:23 -0500823 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000824
David Howellsea5b7782011-01-14 19:10:03 +0000825 switch (err) {
826 case -EBUSY:
827 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500828 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000829 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100830 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000831 path->mnt = mnt;
832 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000833 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500834 default:
835 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000836 }
Al Viro19a167a2011-01-17 01:35:23 -0500837
David Howells9875cf82011-01-14 18:45:21 +0000838}
839
840/*
841 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000842 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000843 * - Flagged as mountpoint
844 * - Flagged as automount point
845 *
846 * This may only be called in refwalk mode.
847 *
848 * Serialization is taken care of in namespace.c
849 */
850static int follow_managed(struct path *path, unsigned flags)
851{
Al Viro8aef1882011-06-16 15:10:06 +0100852 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000853 unsigned managed;
854 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100855 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000856
857 /* Given that we're not holding a lock here, we retain the value in a
858 * local variable for each dentry as we look at it so that we don't see
859 * the components of that value change under us */
860 while (managed = ACCESS_ONCE(path->dentry->d_flags),
861 managed &= DCACHE_MANAGED_DENTRY,
862 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000863 /* Allow the filesystem to manage the transit without i_mutex
864 * being held. */
865 if (managed & DCACHE_MANAGE_TRANSIT) {
866 BUG_ON(!path->dentry->d_op);
867 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400868 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000869 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100870 break;
David Howellscc53ce52011-01-14 18:45:26 +0000871 }
872
David Howells9875cf82011-01-14 18:45:21 +0000873 /* Transit to a mounted filesystem. */
874 if (managed & DCACHE_MOUNTED) {
875 struct vfsmount *mounted = lookup_mnt(path);
876 if (mounted) {
877 dput(path->dentry);
878 if (need_mntput)
879 mntput(path->mnt);
880 path->mnt = mounted;
881 path->dentry = dget(mounted->mnt_root);
882 need_mntput = true;
883 continue;
884 }
885
886 /* Something is mounted on this dentry in another
887 * namespace and/or whatever was mounted there in this
888 * namespace got unmounted before we managed to get the
889 * vfsmount_lock */
890 }
891
892 /* Handle an automount point */
893 if (managed & DCACHE_NEED_AUTOMOUNT) {
894 ret = follow_automount(path, flags, &need_mntput);
895 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100896 break;
David Howells9875cf82011-01-14 18:45:21 +0000897 continue;
898 }
899
900 /* We didn't change the current path point */
901 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
Al Viro8aef1882011-06-16 15:10:06 +0100903
904 if (need_mntput && path->mnt == mnt)
905 mntput(path->mnt);
906 if (ret == -EISDIR)
907 ret = 0;
908 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909}
910
David Howellscc53ce52011-01-14 18:45:26 +0000911int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
913 struct vfsmount *mounted;
914
Al Viro1c755af2009-04-18 14:06:57 -0400915 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400917 dput(path->dentry);
918 mntput(path->mnt);
919 path->mnt = mounted;
920 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 return 1;
922 }
923 return 0;
924}
925
Ian Kent62a73752011-03-25 01:51:02 +0800926static inline bool managed_dentry_might_block(struct dentry *dentry)
927{
928 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
929 dentry->d_op->d_manage(dentry, true) < 0);
930}
931
David Howells9875cf82011-01-14 18:45:21 +0000932/*
Al Viro287548e2011-05-27 06:50:06 -0400933 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
934 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000935 */
936static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400937 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000938{
Ian Kent62a73752011-03-25 01:51:02 +0800939 for (;;) {
David Howells9875cf82011-01-14 18:45:21 +0000940 struct vfsmount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800941 /*
942 * Don't forget we might have a non-mountpoint managed dentry
943 * that wants to block transit.
944 */
Al Viro287548e2011-05-27 06:50:06 -0400945 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000946 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800947
948 if (!d_mountpoint(path->dentry))
949 break;
950
David Howells9875cf82011-01-14 18:45:21 +0000951 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
952 if (!mounted)
953 break;
954 path->mnt = mounted;
955 path->dentry = mounted->mnt_root;
956 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700957 /*
958 * Update the inode too. We don't need to re-check the
959 * dentry sequence number here after this d_inode read,
960 * because a mount-point is always pinned.
961 */
962 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000963 }
David Howells9875cf82011-01-14 18:45:21 +0000964 return true;
965}
966
Al Virodea39372011-05-27 06:53:39 -0400967static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400968{
Al Virodea39372011-05-27 06:53:39 -0400969 while (d_mountpoint(nd->path.dentry)) {
Al Viro287548e2011-05-27 06:50:06 -0400970 struct vfsmount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400971 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400972 if (!mounted)
973 break;
Al Virodea39372011-05-27 06:53:39 -0400974 nd->path.mnt = mounted;
975 nd->path.dentry = mounted->mnt_root;
976 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400977 }
978}
979
Nick Piggin31e6b012011-01-07 17:49:52 +1100980static int follow_dotdot_rcu(struct nameidata *nd)
981{
Nick Piggin31e6b012011-01-07 17:49:52 +1100982 set_root_rcu(nd);
983
David Howells9875cf82011-01-14 18:45:21 +0000984 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100985 if (nd->path.dentry == nd->root.dentry &&
986 nd->path.mnt == nd->root.mnt) {
987 break;
988 }
989 if (nd->path.dentry != nd->path.mnt->mnt_root) {
990 struct dentry *old = nd->path.dentry;
991 struct dentry *parent = old->d_parent;
992 unsigned seq;
993
994 seq = read_seqcount_begin(&parent->d_seq);
995 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500996 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100997 nd->path.dentry = parent;
998 nd->seq = seq;
999 break;
1000 }
1001 if (!follow_up_rcu(&nd->path))
1002 break;
1003 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001004 }
Al Virodea39372011-05-27 06:53:39 -04001005 follow_mount_rcu(nd);
1006 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001007 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001008
1009failed:
1010 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001011 if (!(nd->flags & LOOKUP_ROOT))
1012 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -05001013 rcu_read_unlock();
1014 br_read_unlock(vfsmount_lock);
1015 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001016}
1017
David Howells9875cf82011-01-14 18:45:21 +00001018/*
David Howellscc53ce52011-01-14 18:45:26 +00001019 * Follow down to the covering mount currently visible to userspace. At each
1020 * point, the filesystem owning that dentry may be queried as to whether the
1021 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001022 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001023int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001024{
1025 unsigned managed;
1026 int ret;
1027
1028 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1029 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1030 /* Allow the filesystem to manage the transit without i_mutex
1031 * being held.
1032 *
1033 * We indicate to the filesystem if someone is trying to mount
1034 * something here. This gives autofs the chance to deny anyone
1035 * other than its daemon the right to mount on its
1036 * superstructure.
1037 *
1038 * The filesystem may sleep at this point.
1039 */
1040 if (managed & DCACHE_MANAGE_TRANSIT) {
1041 BUG_ON(!path->dentry->d_op);
1042 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001043 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001044 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001045 if (ret < 0)
1046 return ret == -EISDIR ? 0 : ret;
1047 }
1048
1049 /* Transit to a mounted filesystem. */
1050 if (managed & DCACHE_MOUNTED) {
1051 struct vfsmount *mounted = lookup_mnt(path);
1052 if (!mounted)
1053 break;
1054 dput(path->dentry);
1055 mntput(path->mnt);
1056 path->mnt = mounted;
1057 path->dentry = dget(mounted->mnt_root);
1058 continue;
1059 }
1060
1061 /* Don't handle automount points here */
1062 break;
1063 }
1064 return 0;
1065}
1066
1067/*
David Howells9875cf82011-01-14 18:45:21 +00001068 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1069 */
1070static void follow_mount(struct path *path)
1071{
1072 while (d_mountpoint(path->dentry)) {
1073 struct vfsmount *mounted = lookup_mnt(path);
1074 if (!mounted)
1075 break;
1076 dput(path->dentry);
1077 mntput(path->mnt);
1078 path->mnt = mounted;
1079 path->dentry = dget(mounted->mnt_root);
1080 }
1081}
1082
Nick Piggin31e6b012011-01-07 17:49:52 +11001083static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
Al Viro2a737872009-04-07 11:49:53 -04001085 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001088 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Al Viro2a737872009-04-07 11:49:53 -04001090 if (nd->path.dentry == nd->root.dentry &&
1091 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 break;
1093 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001094 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001095 /* rare case of legitimate dget_parent()... */
1096 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 dput(old);
1098 break;
1099 }
Al Viro3088dd72010-01-30 15:47:29 -05001100 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
Al Viro79ed0222009-04-18 13:59:41 -04001103 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001104 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105}
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001108 * Allocate a dentry with name and parent, and perform a parent
1109 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1110 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1111 * have verified that no child exists while under i_mutex.
1112 */
1113static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1114 struct qstr *name, struct nameidata *nd)
1115{
1116 struct inode *inode = parent->d_inode;
1117 struct dentry *dentry;
1118 struct dentry *old;
1119
1120 /* Don't create child dentry for a dead directory. */
1121 if (unlikely(IS_DEADDIR(inode)))
1122 return ERR_PTR(-ENOENT);
1123
1124 dentry = d_alloc(parent, name);
1125 if (unlikely(!dentry))
1126 return ERR_PTR(-ENOMEM);
1127
1128 old = inode->i_op->lookup(inode, dentry, nd);
1129 if (unlikely(old)) {
1130 dput(dentry);
1131 dentry = old;
1132 }
1133 return dentry;
1134}
1135
1136/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 * It's more convoluted than I'd like it to be, but... it's still fairly
1138 * small and for now I'd prefer to have fast path as straight as possible.
1139 * It _is_ time-critical.
1140 */
1141static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001142 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
Jan Blunck4ac91372008-02-14 19:34:32 -08001144 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001145 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001146 int need_reval = 1;
1147 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001148 int err;
1149
Al Viro3cac2602009-08-13 18:27:43 +04001150 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001151 * Rename seqlock is not required here because in the off chance
1152 * of a false negative due to a concurrent rename, we're going to
1153 * do the non-racy lookup, below.
1154 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001155 if (nd->flags & LOOKUP_RCU) {
1156 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001157 *inode = nd->inode;
1158 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001159 if (!dentry)
1160 goto unlazy;
1161
Nick Piggin31e6b012011-01-07 17:49:52 +11001162 /* Memory barrier in read_seqcount_begin of child is enough */
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001165 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001166
Al Viro24643082011-02-15 01:26:22 -05001167 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001168 status = d_revalidate(dentry, nd);
1169 if (unlikely(status <= 0)) {
1170 if (status != -ECHILD)
1171 need_reval = 0;
1172 goto unlazy;
1173 }
Al Viro24643082011-02-15 01:26:22 -05001174 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001175 path->mnt = mnt;
1176 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001177 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1178 goto unlazy;
1179 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1180 goto unlazy;
1181 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001182unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001183 if (unlazy_walk(nd, dentry))
1184 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001185 } else {
1186 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001187 }
Al Viro5a18fff2011-03-11 04:44:53 -05001188
1189retry:
1190 if (unlikely(!dentry)) {
1191 struct inode *dir = parent->d_inode;
1192 BUG_ON(nd->inode != dir);
1193
1194 mutex_lock(&dir->i_mutex);
1195 dentry = d_lookup(parent, name);
1196 if (likely(!dentry)) {
1197 dentry = d_alloc_and_lookup(parent, name, nd);
1198 if (IS_ERR(dentry)) {
1199 mutex_unlock(&dir->i_mutex);
1200 return PTR_ERR(dentry);
1201 }
1202 /* known good */
1203 need_reval = 0;
1204 status = 1;
1205 }
1206 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001207 }
Al Viro5a18fff2011-03-11 04:44:53 -05001208 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1209 status = d_revalidate(dentry, nd);
1210 if (unlikely(status <= 0)) {
1211 if (status < 0) {
1212 dput(dentry);
1213 return status;
1214 }
1215 if (!d_invalidate(dentry)) {
1216 dput(dentry);
1217 dentry = NULL;
1218 need_reval = 1;
1219 goto retry;
1220 }
1221 }
1222
David Howells9875cf82011-01-14 18:45:21 +00001223 path->mnt = mnt;
1224 path->dentry = dentry;
1225 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001226 if (unlikely(err < 0)) {
1227 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001228 return err;
Ian Kent89312212011-01-18 12:06:10 +08001229 }
David Howells9875cf82011-01-14 18:45:21 +00001230 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
1233
Al Viro52094c82011-02-21 21:34:47 -05001234static inline int may_lookup(struct nameidata *nd)
1235{
1236 if (nd->flags & LOOKUP_RCU) {
1237 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1238 if (err != -ECHILD)
1239 return err;
Al Viro19660af2011-03-25 10:32:48 -04001240 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001241 return -ECHILD;
1242 }
1243 return exec_permission(nd->inode, 0);
1244}
1245
Al Viro9856fa12011-03-04 14:22:06 -05001246static inline int handle_dots(struct nameidata *nd, int type)
1247{
1248 if (type == LAST_DOTDOT) {
1249 if (nd->flags & LOOKUP_RCU) {
1250 if (follow_dotdot_rcu(nd))
1251 return -ECHILD;
1252 } else
1253 follow_dotdot(nd);
1254 }
1255 return 0;
1256}
1257
Al Viro951361f2011-03-04 14:44:37 -05001258static void terminate_walk(struct nameidata *nd)
1259{
1260 if (!(nd->flags & LOOKUP_RCU)) {
1261 path_put(&nd->path);
1262 } else {
1263 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001264 if (!(nd->flags & LOOKUP_ROOT))
1265 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001266 rcu_read_unlock();
1267 br_read_unlock(vfsmount_lock);
1268 }
1269}
1270
Al Viroce57dfc2011-03-13 19:58:58 -04001271static inline int walk_component(struct nameidata *nd, struct path *path,
1272 struct qstr *name, int type, int follow)
1273{
1274 struct inode *inode;
1275 int err;
1276 /*
1277 * "." and ".." are special - ".." especially so because it has
1278 * to be able to know about the current root directory and
1279 * parent relationships.
1280 */
1281 if (unlikely(type != LAST_NORM))
1282 return handle_dots(nd, type);
1283 err = do_lookup(nd, name, path, &inode);
1284 if (unlikely(err)) {
1285 terminate_walk(nd);
1286 return err;
1287 }
1288 if (!inode) {
1289 path_to_nameidata(path, nd);
1290 terminate_walk(nd);
1291 return -ENOENT;
1292 }
1293 if (unlikely(inode->i_op->follow_link) && follow) {
Al Viro19660af2011-03-25 10:32:48 -04001294 if (nd->flags & LOOKUP_RCU) {
1295 if (unlikely(unlazy_walk(nd, path->dentry))) {
1296 terminate_walk(nd);
1297 return -ECHILD;
1298 }
1299 }
Al Viroce57dfc2011-03-13 19:58:58 -04001300 BUG_ON(inode != path->dentry->d_inode);
1301 return 1;
1302 }
1303 path_to_nameidata(path, nd);
1304 nd->inode = inode;
1305 return 0;
1306}
1307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308/*
Al Virob3563792011-03-14 21:54:55 -04001309 * This limits recursive symlink follows to 8, while
1310 * limiting consecutive symlinks to 40.
1311 *
1312 * Without that kind of total limit, nasty chains of consecutive
1313 * symlinks can cause almost arbitrarily long lookups.
1314 */
1315static inline int nested_symlink(struct path *path, struct nameidata *nd)
1316{
1317 int res;
1318
Al Virob3563792011-03-14 21:54:55 -04001319 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1320 path_put_conditional(path, nd);
1321 path_put(&nd->path);
1322 return -ELOOP;
1323 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001324 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001325
1326 nd->depth++;
1327 current->link_count++;
1328
1329 do {
1330 struct path link = *path;
1331 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001332
1333 res = follow_link(&link, nd, &cookie);
Al Virob3563792011-03-14 21:54:55 -04001334 if (!res)
1335 res = walk_component(nd, path, &nd->last,
1336 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001337 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001338 } while (res > 0);
1339
1340 current->link_count--;
1341 nd->depth--;
1342 return res;
1343}
1344
1345/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001347 * This is the basic name resolution function, turning a pathname into
1348 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001350 * Returns 0 and nd will have valid dentry and mnt on success.
1351 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 */
Al Viro6de88d72009-08-09 01:41:57 +04001353static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354{
1355 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 int err;
1357 unsigned int lookup_flags = nd->flags;
1358
1359 while (*name=='/')
1360 name++;
1361 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001362 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 /* At this point we know we have a real path component. */
1365 for(;;) {
1366 unsigned long hash;
1367 struct qstr this;
1368 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001369 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Trond Myklebustcdce5d62005-10-18 14:20:18 -07001371 nd->flags |= LOOKUP_CONTINUE;
Al Viro52094c82011-02-21 21:34:47 -05001372
1373 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (err)
1375 break;
1376
1377 this.name = name;
1378 c = *(const unsigned char *)name;
1379
1380 hash = init_name_hash();
1381 do {
1382 name++;
1383 hash = partial_name_hash(c, hash);
1384 c = *(const unsigned char *)name;
1385 } while (c && (c != '/'));
1386 this.len = name - (const char *) this.name;
1387 this.hash = end_name_hash(hash);
1388
Al Virofe479a52011-02-22 15:10:03 -05001389 type = LAST_NORM;
1390 if (this.name[0] == '.') switch (this.len) {
1391 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001392 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001393 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001394 nd->flags |= LOOKUP_JUMPED;
1395 }
Al Virofe479a52011-02-22 15:10:03 -05001396 break;
1397 case 1:
1398 type = LAST_DOT;
1399 }
Al Viro5a202bc2011-03-08 14:17:44 -05001400 if (likely(type == LAST_NORM)) {
1401 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001402 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001403 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1404 err = parent->d_op->d_hash(parent, nd->inode,
1405 &this);
1406 if (err < 0)
1407 break;
1408 }
1409 }
Al Virofe479a52011-02-22 15:10:03 -05001410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 /* remove trailing slashes? */
1412 if (!c)
1413 goto last_component;
1414 while (*++name == '/');
1415 if (!*name)
Al Virob3563792011-03-14 21:54:55 -04001416 goto last_component;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Al Viroce57dfc2011-03-13 19:58:58 -04001418 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1419 if (err < 0)
1420 return err;
Al Virofe479a52011-02-22 15:10:03 -05001421
Al Viroce57dfc2011-03-13 19:58:58 -04001422 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001423 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001425 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001428 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 break;
1430 continue;
1431 /* here ends the main loop */
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -08001434 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1435 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Al Virob3563792011-03-14 21:54:55 -04001436 nd->last = this;
1437 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001438 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
Al Viro951361f2011-03-04 14:44:37 -05001440 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return err;
1442}
1443
Al Viro70e9b352011-03-05 21:12:22 -05001444static int path_init(int dfd, const char *name, unsigned int flags,
1445 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001447 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001448 int fput_needed;
1449 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
1451 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001452 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001454 if (flags & LOOKUP_ROOT) {
1455 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001456 if (*name) {
1457 if (!inode->i_op->lookup)
1458 return -ENOTDIR;
1459 retval = inode_permission(inode, MAY_EXEC);
1460 if (retval)
1461 return retval;
1462 }
Al Viro5b6ca022011-03-09 23:04:47 -05001463 nd->path = nd->root;
1464 nd->inode = inode;
1465 if (flags & LOOKUP_RCU) {
1466 br_read_lock(vfsmount_lock);
1467 rcu_read_lock();
1468 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1469 } else {
1470 path_get(&nd->path);
1471 }
1472 return 0;
1473 }
1474
Al Viro2a737872009-04-07 11:49:53 -04001475 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001478 if (flags & LOOKUP_RCU) {
1479 br_read_lock(vfsmount_lock);
1480 rcu_read_lock();
1481 set_root_rcu(nd);
1482 } else {
1483 set_root(nd);
1484 path_get(&nd->root);
1485 }
Al Viro2a737872009-04-07 11:49:53 -04001486 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001487 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001488 if (flags & LOOKUP_RCU) {
1489 struct fs_struct *fs = current->fs;
1490 unsigned seq;
1491
1492 br_read_lock(vfsmount_lock);
1493 rcu_read_lock();
1494
1495 do {
1496 seq = read_seqcount_begin(&fs->seq);
1497 nd->path = fs->pwd;
1498 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1499 } while (read_seqcount_retry(&fs->seq, seq));
1500 } else {
1501 get_fs_pwd(current->fs, &nd->path);
1502 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001503 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001504 struct dentry *dentry;
1505
Al Viro1abf0c72011-03-13 03:51:11 -04001506 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001507 retval = -EBADF;
1508 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001509 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001510
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001511 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001512
Al Virof52e0c12011-03-14 18:56:51 -04001513 if (*name) {
1514 retval = -ENOTDIR;
1515 if (!S_ISDIR(dentry->d_inode->i_mode))
1516 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001517
Al Virof52e0c12011-03-14 18:56:51 -04001518 retval = file_permission(file, MAY_EXEC);
1519 if (retval)
1520 goto fput_fail;
1521 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001522
Jan Blunck5dd784d2008-02-14 19:34:38 -08001523 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001524 if (flags & LOOKUP_RCU) {
1525 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001526 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001527 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1528 br_read_lock(vfsmount_lock);
1529 rcu_read_lock();
1530 } else {
1531 path_get(&file->f_path);
1532 fput_light(file, fput_needed);
1533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 }
Al Viroe41f7d42011-02-22 14:02:58 -05001535
Nick Piggin31e6b012011-01-07 17:49:52 +11001536 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001537 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001538
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001539fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001540 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001541out_fail:
1542 return retval;
1543}
1544
Al Virobd92d7f2011-03-14 19:54:59 -04001545static inline int lookup_last(struct nameidata *nd, struct path *path)
1546{
1547 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1548 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1549
1550 nd->flags &= ~LOOKUP_PARENT;
1551 return walk_component(nd, path, &nd->last, nd->last_type,
1552 nd->flags & LOOKUP_FOLLOW);
1553}
1554
Al Viro9b4a9b12009-04-07 11:44:16 -04001555/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001556static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001557 unsigned int flags, struct nameidata *nd)
1558{
Al Viro70e9b352011-03-05 21:12:22 -05001559 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001560 struct path path;
1561 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001562
1563 /*
1564 * Path walking is largely split up into 2 different synchronisation
1565 * schemes, rcu-walk and ref-walk (explained in
1566 * Documentation/filesystems/path-lookup.txt). These share much of the
1567 * path walk code, but some things particularly setup, cleanup, and
1568 * following mounts are sufficiently divergent that functions are
1569 * duplicated. Typically there is a function foo(), and its RCU
1570 * analogue, foo_rcu().
1571 *
1572 * -ECHILD is the error number of choice (just to avoid clashes) that
1573 * is returned if some aspect of an rcu-walk fails. Such an error must
1574 * be handled by restarting a traditional ref-walk (which will always
1575 * be able to complete).
1576 */
Al Virobd92d7f2011-03-14 19:54:59 -04001577 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001578
Al Virobd92d7f2011-03-14 19:54:59 -04001579 if (unlikely(err))
1580 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001581
1582 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001583 err = link_path_walk(name, nd);
1584
1585 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001586 err = lookup_last(nd, &path);
1587 while (err > 0) {
1588 void *cookie;
1589 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001590 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001591 err = follow_link(&link, nd, &cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001592 if (!err)
1593 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001594 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001595 }
1596 }
Al Viroee0827c2011-02-21 23:38:09 -05001597
Al Viro9f1fafe2011-03-25 11:00:12 -04001598 if (!err)
1599 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001600
1601 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1602 if (!nd->inode->i_op->lookup) {
1603 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001604 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001605 }
1606 }
Al Viro16c2cd72011-02-22 15:50:10 -05001607
Al Viro70e9b352011-03-05 21:12:22 -05001608 if (base)
1609 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001610
Al Viro5b6ca022011-03-09 23:04:47 -05001611 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001612 path_put(&nd->root);
1613 nd->root.mnt = NULL;
1614 }
Al Virobd92d7f2011-03-14 19:54:59 -04001615 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001616}
Nick Piggin31e6b012011-01-07 17:49:52 +11001617
Al Viroee0827c2011-02-21 23:38:09 -05001618static int do_path_lookup(int dfd, const char *name,
1619 unsigned int flags, struct nameidata *nd)
1620{
1621 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1622 if (unlikely(retval == -ECHILD))
1623 retval = path_lookupat(dfd, name, flags, nd);
1624 if (unlikely(retval == -ESTALE))
1625 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001626
1627 if (likely(!retval)) {
1628 if (unlikely(!audit_dummy_context())) {
1629 if (nd->path.dentry && nd->inode)
1630 audit_inode(name, nd->path.dentry);
1631 }
1632 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001633 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634}
1635
Al Viroc9c6cac2011-02-16 15:15:47 -05001636int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001637{
Al Viroc9c6cac2011-02-16 15:15:47 -05001638 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001639}
1640
Al Virod1811462008-08-02 00:49:18 -04001641int kern_path(const char *name, unsigned int flags, struct path *path)
1642{
1643 struct nameidata nd;
1644 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1645 if (!res)
1646 *path = nd.path;
1647 return res;
1648}
1649
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001650/**
1651 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1652 * @dentry: pointer to dentry of the base directory
1653 * @mnt: pointer to vfs mount of the base directory
1654 * @name: pointer to file name
1655 * @flags: lookup flags
1656 * @nd: pointer to nameidata
1657 */
1658int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1659 const char *name, unsigned int flags,
1660 struct nameidata *nd)
1661{
Al Viro5b6ca022011-03-09 23:04:47 -05001662 nd->root.dentry = dentry;
1663 nd->root.mnt = mnt;
1664 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1665 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001666}
1667
Christoph Hellwigeead1912007-10-16 23:25:38 -07001668static struct dentry *__lookup_hash(struct qstr *name,
1669 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001671 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001672 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 int err;
1674
Nick Pigginb74c79e2011-01-07 17:49:58 +11001675 err = exec_permission(inode, 0);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001676 if (err)
1677 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001680 * Don't bother with __d_lookup: callers are for creat as
1681 * well as unlink, so a lot of the time it would cost
1682 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001683 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001684 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001685
Nick Pigginfb045ad2011-01-07 17:49:55 +11001686 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro6e6b1bd2009-08-13 23:38:37 +04001687 dentry = do_revalidate(dentry, nd);
1688
Nick Pigginbaa03892010-08-18 04:37:31 +10001689 if (!dentry)
1690 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 return dentry;
1693}
1694
James Morris057f6c02007-04-26 00:12:05 -07001695/*
1696 * Restricted form of lookup. Doesn't follow links, single-component only,
1697 * needs parent already locked. Doesn't follow mounts.
1698 * SMP-safe.
1699 */
Adrian Bunka244e162006-03-31 02:32:11 -08001700static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
Jan Blunck4ac91372008-02-14 19:34:32 -08001702 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703}
1704
Christoph Hellwigeead1912007-10-16 23:25:38 -07001705/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001706 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001707 * @name: pathname component to lookup
1708 * @base: base directory to lookup from
1709 * @len: maximum length @len should be interpreted to
1710 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001711 * Note that this routine is purely a helper for filesystem usage and should
1712 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001713 * nameidata argument is passed to the filesystem methods and a filesystem
1714 * using this helper needs to be prepared for that.
1715 */
James Morris057f6c02007-04-26 00:12:05 -07001716struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1717{
James Morris057f6c02007-04-26 00:12:05 -07001718 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001719 unsigned long hash;
1720 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001721
David Woodhouse2f9092e2009-04-20 23:18:37 +01001722 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1723
Al Viro6a96ba52011-03-07 23:49:20 -05001724 this.name = name;
1725 this.len = len;
1726 if (!len)
1727 return ERR_PTR(-EACCES);
1728
1729 hash = init_name_hash();
1730 while (len--) {
1731 c = *(const unsigned char *)name++;
1732 if (c == '/' || c == '\0')
1733 return ERR_PTR(-EACCES);
1734 hash = partial_name_hash(c, hash);
1735 }
1736 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001737 /*
1738 * See if the low-level filesystem might want
1739 * to use its own hash..
1740 */
1741 if (base->d_flags & DCACHE_OP_HASH) {
1742 int err = base->d_op->d_hash(base, base->d_inode, &this);
1743 if (err < 0)
1744 return ERR_PTR(err);
1745 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001746
Christoph Hellwig49705b72005-11-08 21:35:06 -08001747 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001748}
1749
Al Viro2d8f3032008-07-22 09:59:21 -04001750int user_path_at(int dfd, const char __user *name, unsigned flags,
1751 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
Al Viro2d8f3032008-07-22 09:59:21 -04001753 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001754 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001757
1758 BUG_ON(flags & LOOKUP_PARENT);
1759
1760 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001762 if (!err)
1763 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 }
1765 return err;
1766}
1767
Al Viro2ad94ae2008-07-21 09:32:51 -04001768static int user_path_parent(int dfd, const char __user *path,
1769 struct nameidata *nd, char **name)
1770{
1771 char *s = getname(path);
1772 int error;
1773
1774 if (IS_ERR(s))
1775 return PTR_ERR(s);
1776
1777 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1778 if (error)
1779 putname(s);
1780 else
1781 *name = s;
1782
1783 return error;
1784}
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786/*
1787 * It's inline, so penalty for filesystems that don't use sticky bit is
1788 * minimal.
1789 */
1790static inline int check_sticky(struct inode *dir, struct inode *inode)
1791{
David Howellsda9592e2008-11-14 10:39:05 +11001792 uid_t fsuid = current_fsuid();
1793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 if (!(dir->i_mode & S_ISVTX))
1795 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001796 if (current_user_ns() != inode_userns(inode))
1797 goto other_userns;
David Howellsda9592e2008-11-14 10:39:05 +11001798 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001800 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 return 0;
Serge E. Hallyne795b712011-03-23 16:43:25 -07001802
1803other_userns:
1804 return !ns_capable(inode_userns(inode), CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805}
1806
1807/*
1808 * Check whether we can remove a link victim from directory dir, check
1809 * whether the type of victim is right.
1810 * 1. We can't do it if dir is read-only (done in permission())
1811 * 2. We should have write and exec permissions on dir
1812 * 3. We can't remove anything from append-only dir
1813 * 4. We can't do anything with immutable dir (done in permission())
1814 * 5. If the sticky bit on dir is set we should either
1815 * a. be owner of dir, or
1816 * b. be owner of victim, or
1817 * c. have CAP_FOWNER capability
1818 * 6. If the victim is append-only or immutable we can't do antyhing with
1819 * links pointing to it.
1820 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1821 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1822 * 9. We can't remove a root or mountpoint.
1823 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1824 * nfs_async_unlink().
1825 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001826static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827{
1828 int error;
1829
1830 if (!victim->d_inode)
1831 return -ENOENT;
1832
1833 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001834 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Al Virof419a2e2008-07-22 00:07:17 -04001836 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 if (error)
1838 return error;
1839 if (IS_APPEND(dir))
1840 return -EPERM;
1841 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001842 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 return -EPERM;
1844 if (isdir) {
1845 if (!S_ISDIR(victim->d_inode->i_mode))
1846 return -ENOTDIR;
1847 if (IS_ROOT(victim))
1848 return -EBUSY;
1849 } else if (S_ISDIR(victim->d_inode->i_mode))
1850 return -EISDIR;
1851 if (IS_DEADDIR(dir))
1852 return -ENOENT;
1853 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1854 return -EBUSY;
1855 return 0;
1856}
1857
1858/* Check whether we can create an object with dentry child in directory
1859 * dir.
1860 * 1. We can't do it if child already exists (open has special treatment for
1861 * this case, but since we are inlined it's OK)
1862 * 2. We can't do it if dir is read-only (done in permission())
1863 * 3. We should have write and exec permissions on dir
1864 * 4. We can't do it if dir is immutable (done in permission())
1865 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001866static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867{
1868 if (child->d_inode)
1869 return -EEXIST;
1870 if (IS_DEADDIR(dir))
1871 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001872 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873}
1874
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875/*
1876 * p1 and p2 should be directories on the same fs.
1877 */
1878struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1879{
1880 struct dentry *p;
1881
1882 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001883 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 return NULL;
1885 }
1886
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001887 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001889 p = d_ancestor(p2, p1);
1890 if (p) {
1891 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1892 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1893 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
1895
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001896 p = d_ancestor(p1, p2);
1897 if (p) {
1898 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1899 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1900 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 }
1902
Ingo Molnarf2eace22006-07-03 00:25:05 -07001903 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1904 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 return NULL;
1906}
1907
1908void unlock_rename(struct dentry *p1, struct dentry *p2)
1909{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001910 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001912 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001913 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
1915}
1916
1917int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1918 struct nameidata *nd)
1919{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001920 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
1922 if (error)
1923 return error;
1924
Al Viroacfa4382008-12-04 10:06:33 -05001925 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 return -EACCES; /* shouldn't it be ENOSYS? */
1927 mode &= S_IALLUGO;
1928 mode |= S_IFREG;
1929 error = security_inode_create(dir, dentry, mode);
1930 if (error)
1931 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001933 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001934 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return error;
1936}
1937
Al Viro73d049a2011-03-11 12:08:24 -05001938static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001940 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 struct inode *inode = dentry->d_inode;
1942 int error;
1943
Al Virobcda7652011-03-13 16:42:14 -04001944 /* O_PATH? */
1945 if (!acc_mode)
1946 return 0;
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 if (!inode)
1949 return -ENOENT;
1950
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001951 switch (inode->i_mode & S_IFMT) {
1952 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001954 case S_IFDIR:
1955 if (acc_mode & MAY_WRITE)
1956 return -EISDIR;
1957 break;
1958 case S_IFBLK:
1959 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001960 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001962 /*FALLTHRU*/
1963 case S_IFIFO:
1964 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001966 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001967 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001968
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001969 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001970 if (error)
1971 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 /*
1974 * An append-only file must be opened in append mode for writing.
1975 */
1976 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001977 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001978 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001980 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
1982
1983 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001984 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05001985 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987 /*
1988 * Ensure there are no outstanding leases on the file.
1989 */
Al Virob65a9cf2009-12-16 06:27:40 -05001990 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001991}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
Jeff Laytone1181ee2010-12-07 16:19:50 -05001993static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05001994{
Jeff Laytone1181ee2010-12-07 16:19:50 -05001995 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05001996 struct inode *inode = path->dentry->d_inode;
1997 int error = get_write_access(inode);
1998 if (error)
1999 return error;
2000 /*
2001 * Refuse to truncate files with mandatory locks held on them.
2002 */
2003 error = locks_verify_locked(inode);
2004 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002005 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002006 if (!error) {
2007 error = do_truncate(path->dentry, 0,
2008 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002009 filp);
Al Viro7715b522009-12-16 03:54:00 -05002010 }
2011 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002012 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013}
2014
Dave Hansend57999e2008-02-15 14:37:27 -08002015/*
Dave Hansend57999e2008-02-15 14:37:27 -08002016 * Note that while the flag value (low two bits) for sys_open means:
2017 * 00 - read-only
2018 * 01 - write-only
2019 * 10 - read-write
2020 * 11 - special
2021 * it is changed into
2022 * 00 - no permissions needed
2023 * 01 - read-permission
2024 * 10 - write-permission
2025 * 11 - read-write
2026 * for the internal routines (ie open_namei()/follow_link() etc)
2027 * This is more logical, and also allows the 00 "no perm needed"
2028 * to be used for symlinks (where the permissions are checked
2029 * later).
2030 *
2031*/
2032static inline int open_to_namei_flags(int flag)
2033{
2034 if ((flag+1) & O_ACCMODE)
2035 flag++;
2036 return flag;
2037}
2038
Nick Piggin31e6b012011-01-07 17:49:52 +11002039/*
Al Virofe2d35f2011-03-05 22:58:25 -05002040 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002041 */
Al Virofb1cc552009-12-24 01:58:28 -05002042static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002043 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002044{
Al Viroa1e28032009-12-24 02:12:06 -05002045 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002046 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002047 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002048 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002049 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002050 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002051 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002052 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002053
Al Viroc3e380b2011-02-23 13:39:45 -05002054 nd->flags &= ~LOOKUP_PARENT;
2055 nd->flags |= op->intent;
2056
Al Viro1f36f772009-12-26 10:56:19 -05002057 switch (nd->last_type) {
2058 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002059 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002060 error = handle_dots(nd, nd->last_type);
2061 if (error)
2062 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002063 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002064 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002065 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002066 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002067 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002068 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002069 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002070 error = -EISDIR;
2071 goto exit;
2072 }
2073 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002074 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002075 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002076 if (error)
Al Viro9f1fafe2011-03-25 11:00:12 -04002077 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002078 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002079 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002080 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002081
Al Viroca344a892011-03-09 00:36:45 -05002082 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002083 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002084 if (nd->last.name[nd->last.len])
2085 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002086 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2087 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002088 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002089 error = walk_component(nd, path, &nd->last, LAST_NORM,
2090 !symlink_ok);
2091 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002092 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002093 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002094 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002095 /* sayonara */
Al Viro9f1fafe2011-03-25 11:00:12 -04002096 error = complete_walk(nd);
2097 if (error)
2098 return ERR_PTR(-ECHILD);
Al Virofe2d35f2011-03-05 22:58:25 -05002099
2100 error = -ENOTDIR;
2101 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002102 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002103 goto exit;
2104 }
2105 audit_inode(pathname, nd->path.dentry);
2106 goto ok;
2107 }
2108
2109 /* create side of things */
Al Viro9f1fafe2011-03-25 11:00:12 -04002110 error = complete_walk(nd);
2111 if (error)
2112 return ERR_PTR(error);
Al Virofe2d35f2011-03-05 22:58:25 -05002113
2114 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002115 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002116 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002117 if (nd->last.name[nd->last.len])
2118 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002119
Al Viroa1e28032009-12-24 02:12:06 -05002120 mutex_lock(&dir->d_inode->i_mutex);
2121
Al Viro6c0d46c2011-03-09 00:59:59 -05002122 dentry = lookup_hash(nd);
2123 error = PTR_ERR(dentry);
2124 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002125 mutex_unlock(&dir->d_inode->i_mutex);
2126 goto exit;
2127 }
2128
Al Viro6c0d46c2011-03-09 00:59:59 -05002129 path->dentry = dentry;
2130 path->mnt = nd->path.mnt;
2131
Al Virofb1cc552009-12-24 01:58:28 -05002132 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002133 if (!dentry->d_inode) {
2134 int mode = op->mode;
2135 if (!IS_POSIXACL(dir->d_inode))
2136 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002137 /*
2138 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002139 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002140 * the time when the file is created and when
2141 * a permanent write count is taken through
2142 * the 'struct file' in nameidata_to_filp().
2143 */
2144 error = mnt_want_write(nd->path.mnt);
2145 if (error)
2146 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002147 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002148 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002149 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002150 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002151 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002152 error = security_path_mknod(&nd->path, dentry, mode, 0);
2153 if (error)
2154 goto exit_mutex_unlock;
2155 error = vfs_create(dir->d_inode, dentry, mode, nd);
2156 if (error)
2157 goto exit_mutex_unlock;
2158 mutex_unlock(&dir->d_inode->i_mutex);
2159 dput(nd->path.dentry);
2160 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002161 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002162 }
2163
2164 /*
2165 * It already exists.
2166 */
2167 mutex_unlock(&dir->d_inode->i_mutex);
2168 audit_inode(pathname, path->dentry);
2169
2170 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002171 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002172 goto exit_dput;
2173
David Howells9875cf82011-01-14 18:45:21 +00002174 error = follow_managed(path, nd->flags);
2175 if (error < 0)
2176 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002177
2178 error = -ENOENT;
2179 if (!path->dentry->d_inode)
2180 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002181
2182 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002183 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002184
2185 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002186 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002187 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002188 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002189 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002190ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002191 if (!S_ISREG(nd->inode->i_mode))
2192 will_truncate = 0;
2193
Al Viro0f9d1a12011-03-09 00:13:14 -05002194 if (will_truncate) {
2195 error = mnt_want_write(nd->path.mnt);
2196 if (error)
2197 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002198 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002199 }
Al Viroca344a892011-03-09 00:36:45 -05002200common:
Al Virobcda7652011-03-13 16:42:14 -04002201 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002202 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002203 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002204 filp = nameidata_to_filp(nd);
2205 if (!IS_ERR(filp)) {
2206 error = ima_file_check(filp, op->acc_mode);
2207 if (error) {
2208 fput(filp);
2209 filp = ERR_PTR(error);
2210 }
2211 }
2212 if (!IS_ERR(filp)) {
2213 if (will_truncate) {
2214 error = handle_truncate(filp);
2215 if (error) {
2216 fput(filp);
2217 filp = ERR_PTR(error);
2218 }
2219 }
2220 }
Al Viroca344a892011-03-09 00:36:45 -05002221out:
2222 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002223 mnt_drop_write(nd->path.mnt);
2224 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002225 return filp;
2226
2227exit_mutex_unlock:
2228 mutex_unlock(&dir->d_inode->i_mutex);
2229exit_dput:
2230 path_put_conditional(path, nd);
2231exit:
Al Viroca344a892011-03-09 00:36:45 -05002232 filp = ERR_PTR(error);
2233 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002234}
2235
Al Viro13aab422011-02-23 17:54:08 -05002236static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002237 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238{
Al Virofe2d35f2011-03-05 22:58:25 -05002239 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002240 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002241 struct path path;
Al Viro13aab422011-02-23 17:54:08 -05002242 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002243
2244 filp = get_empty_filp();
2245 if (!filp)
2246 return ERR_PTR(-ENFILE);
2247
Al Viro47c805d2011-02-23 17:44:09 -05002248 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002249 nd->intent.open.file = filp;
2250 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2251 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002252
Al Viro73d049a2011-03-11 12:08:24 -05002253 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002254 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002255 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002256
Al Virofe2d35f2011-03-05 22:58:25 -05002257 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002258 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002259 if (unlikely(error))
2260 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
Al Viro73d049a2011-03-11 12:08:24 -05002262 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002263 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002264 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002265 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002266 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002267 path_put_conditional(&path, nd);
2268 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002269 filp = ERR_PTR(-ELOOP);
2270 break;
2271 }
Al Viro73d049a2011-03-11 12:08:24 -05002272 nd->flags |= LOOKUP_PARENT;
2273 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002274 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002275 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002276 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002277 else
Al Viro73d049a2011-03-11 12:08:24 -05002278 filp = do_last(nd, &path, op, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002279 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002280 }
Al Viro10fa8e62009-12-26 07:09:49 -05002281out:
Al Viro73d049a2011-03-11 12:08:24 -05002282 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2283 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002284 if (base)
2285 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002286 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002287 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288
Nick Piggin31e6b012011-01-07 17:49:52 +11002289out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002290 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002291 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292}
2293
Al Viro13aab422011-02-23 17:54:08 -05002294struct file *do_filp_open(int dfd, const char *pathname,
2295 const struct open_flags *op, int flags)
2296{
Al Viro73d049a2011-03-11 12:08:24 -05002297 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002298 struct file *filp;
2299
Al Viro73d049a2011-03-11 12:08:24 -05002300 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002301 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002302 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002303 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002304 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002305 return filp;
2306}
2307
Al Viro73d049a2011-03-11 12:08:24 -05002308struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2309 const char *name, const struct open_flags *op, int flags)
2310{
2311 struct nameidata nd;
2312 struct file *file;
2313
2314 nd.root.mnt = mnt;
2315 nd.root.dentry = dentry;
2316
2317 flags |= LOOKUP_ROOT;
2318
Al Virobcda7652011-03-13 16:42:14 -04002319 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002320 return ERR_PTR(-ELOOP);
2321
2322 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2323 if (unlikely(file == ERR_PTR(-ECHILD)))
2324 file = path_openat(-1, name, &nd, op, flags);
2325 if (unlikely(file == ERR_PTR(-ESTALE)))
2326 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2327 return file;
2328}
2329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330/**
2331 * lookup_create - lookup a dentry, creating it if it doesn't exist
2332 * @nd: nameidata info
2333 * @is_dir: directory flag
2334 *
2335 * Simple function to lookup and return a dentry and create it
2336 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002337 *
Jan Blunck4ac91372008-02-14 19:34:32 -08002338 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 */
2340struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2341{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002342 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
Jan Blunck4ac91372008-02-14 19:34:32 -08002344 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002345 /*
2346 * Yucky last component or no last component at all?
2347 * (foo/., foo/.., /////)
2348 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (nd->last_type != LAST_NORM)
2350 goto fail;
2351 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04002352 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04002353 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002354
2355 /*
2356 * Do the final lookup.
2357 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08002358 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 if (IS_ERR(dentry))
2360 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002361
Al Viroe9baf6e2008-05-15 04:49:12 -04002362 if (dentry->d_inode)
2363 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002364 /*
2365 * Special case - lookup gave negative, but... we had foo/bar/
2366 * From the vfs_mknod() POV we just have a negative dentry -
2367 * all is fine. Let's be bastards - you had / on the end, you've
2368 * been asking for (non-existent) directory. -ENOENT for you.
2369 */
Al Viroe9baf6e2008-05-15 04:49:12 -04002370 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2371 dput(dentry);
2372 dentry = ERR_PTR(-ENOENT);
2373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002375eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002377 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378fail:
2379 return dentry;
2380}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07002381EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
2383int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2384{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002385 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
2387 if (error)
2388 return error;
2389
Serge E. Hallyne795b712011-03-23 16:43:25 -07002390 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2391 !ns_capable(inode_userns(dir), CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 return -EPERM;
2393
Al Viroacfa4382008-12-04 10:06:33 -05002394 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return -EPERM;
2396
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002397 error = devcgroup_inode_mknod(mode, dev);
2398 if (error)
2399 return error;
2400
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 error = security_inode_mknod(dir, dentry, mode, dev);
2402 if (error)
2403 return error;
2404
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002406 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002407 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 return error;
2409}
2410
Dave Hansen463c3192008-02-15 14:37:57 -08002411static int may_mknod(mode_t mode)
2412{
2413 switch (mode & S_IFMT) {
2414 case S_IFREG:
2415 case S_IFCHR:
2416 case S_IFBLK:
2417 case S_IFIFO:
2418 case S_IFSOCK:
2419 case 0: /* zero mode translates to S_IFREG */
2420 return 0;
2421 case S_IFDIR:
2422 return -EPERM;
2423 default:
2424 return -EINVAL;
2425 }
2426}
2427
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002428SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2429 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430{
Al Viro2ad94ae2008-07-21 09:32:51 -04002431 int error;
2432 char *tmp;
2433 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 struct nameidata nd;
2435
2436 if (S_ISDIR(mode))
2437 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Al Viro2ad94ae2008-07-21 09:32:51 -04002439 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002441 return error;
2442
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002444 if (IS_ERR(dentry)) {
2445 error = PTR_ERR(dentry);
2446 goto out_unlock;
2447 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002448 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002449 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002450 error = may_mknod(mode);
2451 if (error)
2452 goto out_dput;
2453 error = mnt_want_write(nd.path.mnt);
2454 if (error)
2455 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002456 error = security_path_mknod(&nd.path, dentry, mode, dev);
2457 if (error)
2458 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002459 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002461 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 break;
2463 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002464 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 new_decode_dev(dev));
2466 break;
2467 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002468 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002471out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002472 mnt_drop_write(nd.path.mnt);
2473out_dput:
2474 dput(dentry);
2475out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002476 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002477 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 putname(tmp);
2479
2480 return error;
2481}
2482
Heiko Carstens3480b252009-01-14 14:14:16 +01002483SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002484{
2485 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2486}
2487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2489{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002490 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
2492 if (error)
2493 return error;
2494
Al Viroacfa4382008-12-04 10:06:33 -05002495 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 return -EPERM;
2497
2498 mode &= (S_IRWXUGO|S_ISVTX);
2499 error = security_inode_mkdir(dir, dentry, mode);
2500 if (error)
2501 return error;
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002504 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002505 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 return error;
2507}
2508
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002509SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510{
2511 int error = 0;
2512 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002513 struct dentry *dentry;
2514 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515
Al Viro2ad94ae2008-07-21 09:32:51 -04002516 error = user_path_parent(dfd, pathname, &nd, &tmp);
2517 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002518 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Dave Hansen6902d922006-09-30 23:29:01 -07002520 dentry = lookup_create(&nd, 1);
2521 error = PTR_ERR(dentry);
2522 if (IS_ERR(dentry))
2523 goto out_unlock;
2524
Jan Blunck4ac91372008-02-14 19:34:32 -08002525 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002526 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002527 error = mnt_want_write(nd.path.mnt);
2528 if (error)
2529 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002530 error = security_path_mkdir(&nd.path, dentry, mode);
2531 if (error)
2532 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002533 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002534out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002535 mnt_drop_write(nd.path.mnt);
2536out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002537 dput(dentry);
2538out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002539 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002540 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002541 putname(tmp);
2542out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 return error;
2544}
2545
Heiko Carstens3cdad422009-01-14 14:14:22 +01002546SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002547{
2548 return sys_mkdirat(AT_FDCWD, pathname, mode);
2549}
2550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551/*
Sage Weila71905f2011-05-24 13:06:08 -07002552 * The dentry_unhash() helper will try to drop the dentry early: we
2553 * should have a usage count of 2 if we're the only user of this
2554 * dentry, and if that is true (possibly after pruning the dcache),
2555 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 *
2557 * A low-level filesystem can, if it choses, legally
2558 * do a
2559 *
2560 * if (!d_unhashed(dentry))
2561 * return -EBUSY;
2562 *
2563 * if it cannot handle the case of removing a directory
2564 * that is still in use by something else..
2565 */
2566void dentry_unhash(struct dentry *dentry)
2567{
Vasily Averindc168422006-12-06 20:37:07 -08002568 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07002570 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 __d_drop(dentry);
2572 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573}
2574
2575int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2576{
2577 int error = may_delete(dir, dentry, 1);
2578
2579 if (error)
2580 return error;
2581
Al Viroacfa4382008-12-04 10:06:33 -05002582 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 return -EPERM;
2584
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002585 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Sage Weil912dbc12011-05-24 13:06:11 -07002587 error = -EBUSY;
2588 if (d_mountpoint(dentry))
2589 goto out;
2590
2591 error = security_inode_rmdir(dir, dentry);
2592 if (error)
2593 goto out;
2594
Sage Weil3cebde22011-05-29 21:20:59 -07002595 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07002596 error = dir->i_op->rmdir(dir, dentry);
2597 if (error)
2598 goto out;
2599
2600 dentry->d_inode->i_flags |= S_DEAD;
2601 dont_mount(dentry);
2602
2603out:
2604 mutex_unlock(&dentry->d_inode->i_mutex);
2605 if (!error)
2606 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 return error;
2608}
2609
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002610static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611{
2612 int error = 0;
2613 char * name;
2614 struct dentry *dentry;
2615 struct nameidata nd;
2616
Al Viro2ad94ae2008-07-21 09:32:51 -04002617 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002619 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
2621 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002622 case LAST_DOTDOT:
2623 error = -ENOTEMPTY;
2624 goto exit1;
2625 case LAST_DOT:
2626 error = -EINVAL;
2627 goto exit1;
2628 case LAST_ROOT:
2629 error = -EBUSY;
2630 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002632
2633 nd.flags &= ~LOOKUP_PARENT;
2634
Jan Blunck4ac91372008-02-14 19:34:32 -08002635 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002636 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002638 if (IS_ERR(dentry))
2639 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002640 if (!dentry->d_inode) {
2641 error = -ENOENT;
2642 goto exit3;
2643 }
Dave Hansen06227532008-02-15 14:37:34 -08002644 error = mnt_want_write(nd.path.mnt);
2645 if (error)
2646 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002647 error = security_path_rmdir(&nd.path, dentry);
2648 if (error)
2649 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002650 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002651exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002652 mnt_drop_write(nd.path.mnt);
2653exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002654 dput(dentry);
2655exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002656 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002658 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 putname(name);
2660 return error;
2661}
2662
Heiko Carstens3cdad422009-01-14 14:14:22 +01002663SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002664{
2665 return do_rmdir(AT_FDCWD, pathname);
2666}
2667
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668int vfs_unlink(struct inode *dir, struct dentry *dentry)
2669{
2670 int error = may_delete(dir, dentry, 0);
2671
2672 if (error)
2673 return error;
2674
Al Viroacfa4382008-12-04 10:06:33 -05002675 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 return -EPERM;
2677
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002678 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 if (d_mountpoint(dentry))
2680 error = -EBUSY;
2681 else {
2682 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002683 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002685 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002686 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002689 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
2691 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2692 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002693 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 }
Robert Love0eeca282005-07-12 17:06:03 -04002696
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 return error;
2698}
2699
2700/*
2701 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002702 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 * writeout happening, and we don't want to prevent access to the directory
2704 * while waiting on the I/O.
2705 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002706static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707{
Al Viro2ad94ae2008-07-21 09:32:51 -04002708 int error;
2709 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 struct dentry *dentry;
2711 struct nameidata nd;
2712 struct inode *inode = NULL;
2713
Al Viro2ad94ae2008-07-21 09:32:51 -04002714 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002716 return error;
2717
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 error = -EISDIR;
2719 if (nd.last_type != LAST_NORM)
2720 goto exit1;
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);
2727 if (!IS_ERR(dentry)) {
2728 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03002729 if (nd.last.name[nd.last.len])
2730 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03002732 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04002733 goto slashes;
2734 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002735 error = mnt_want_write(nd.path.mnt);
2736 if (error)
2737 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002738 error = security_path_unlink(&nd.path, dentry);
2739 if (error)
2740 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002741 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002742exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002743 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 exit2:
2745 dput(dentry);
2746 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002747 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 if (inode)
2749 iput(inode); /* truncate the inode here */
2750exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002751 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 putname(name);
2753 return error;
2754
2755slashes:
2756 error = !dentry->d_inode ? -ENOENT :
2757 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2758 goto exit2;
2759}
2760
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002761SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002762{
2763 if ((flag & ~AT_REMOVEDIR) != 0)
2764 return -EINVAL;
2765
2766 if (flag & AT_REMOVEDIR)
2767 return do_rmdir(dfd, pathname);
2768
2769 return do_unlinkat(dfd, pathname);
2770}
2771
Heiko Carstens3480b252009-01-14 14:14:16 +01002772SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002773{
2774 return do_unlinkat(AT_FDCWD, pathname);
2775}
2776
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002777int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002779 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
2781 if (error)
2782 return error;
2783
Al Viroacfa4382008-12-04 10:06:33 -05002784 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 return -EPERM;
2786
2787 error = security_inode_symlink(dir, dentry, oldname);
2788 if (error)
2789 return error;
2790
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002792 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002793 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 return error;
2795}
2796
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002797SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2798 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799{
Al Viro2ad94ae2008-07-21 09:32:51 -04002800 int error;
2801 char *from;
2802 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002803 struct dentry *dentry;
2804 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805
2806 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002807 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002809
2810 error = user_path_parent(newdfd, newname, &nd, &to);
2811 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002812 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
Dave Hansen6902d922006-09-30 23:29:01 -07002814 dentry = lookup_create(&nd, 0);
2815 error = PTR_ERR(dentry);
2816 if (IS_ERR(dentry))
2817 goto out_unlock;
2818
Dave Hansen75c3f292008-02-15 14:37:45 -08002819 error = mnt_want_write(nd.path.mnt);
2820 if (error)
2821 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002822 error = security_path_symlink(&nd.path, dentry, from);
2823 if (error)
2824 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002825 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002826out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002827 mnt_drop_write(nd.path.mnt);
2828out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002829 dput(dentry);
2830out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002831 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002832 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002833 putname(to);
2834out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 putname(from);
2836 return error;
2837}
2838
Heiko Carstens3480b252009-01-14 14:14:16 +01002839SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002840{
2841 return sys_symlinkat(oldname, AT_FDCWD, newname);
2842}
2843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2845{
2846 struct inode *inode = old_dentry->d_inode;
2847 int error;
2848
2849 if (!inode)
2850 return -ENOENT;
2851
Miklos Szeredia95164d2008-07-30 15:08:48 +02002852 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 if (error)
2854 return error;
2855
2856 if (dir->i_sb != inode->i_sb)
2857 return -EXDEV;
2858
2859 /*
2860 * A link to an append-only or immutable file cannot be created.
2861 */
2862 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2863 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002864 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002866 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 return -EPERM;
2868
2869 error = security_inode_link(old_dentry, dir, new_dentry);
2870 if (error)
2871 return error;
2872
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002873 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302874 /* Make sure we don't allow creating hardlink to an unlinked file */
2875 if (inode->i_nlink == 0)
2876 error = -ENOENT;
2877 else
2878 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002879 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002880 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002881 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 return error;
2883}
2884
2885/*
2886 * Hardlinks are often used in delicate situations. We avoid
2887 * security-related surprises by not following symlinks on the
2888 * newname. --KAB
2889 *
2890 * We don't follow them on the oldname either to be compatible
2891 * with linux 2.0, and to avoid hard-linking to directories
2892 * and other special files. --ADM
2893 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002894SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2895 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896{
2897 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002898 struct nameidata nd;
2899 struct path old_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302900 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002902 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302904 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002905 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302906 /*
2907 * To use null names we require CAP_DAC_READ_SEARCH
2908 * This ensures that not everyone will be able to create
2909 * handlink using the passed filedescriptor.
2910 */
2911 if (flags & AT_EMPTY_PATH) {
2912 if (!capable(CAP_DAC_READ_SEARCH))
2913 return -ENOENT;
2914 how = LOOKUP_EMPTY;
2915 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002916
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302917 if (flags & AT_SYMLINK_FOLLOW)
2918 how |= LOOKUP_FOLLOW;
2919
2920 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002922 return error;
2923
2924 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 if (error)
2926 goto out;
2927 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002928 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 goto out_release;
2930 new_dentry = lookup_create(&nd, 0);
2931 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002932 if (IS_ERR(new_dentry))
2933 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002934 error = mnt_want_write(nd.path.mnt);
2935 if (error)
2936 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002937 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2938 if (error)
2939 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002940 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002941out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002942 mnt_drop_write(nd.path.mnt);
2943out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002944 dput(new_dentry);
2945out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002946 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002948 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002949 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950out:
Al Viro2d8f3032008-07-22 09:59:21 -04002951 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952
2953 return error;
2954}
2955
Heiko Carstens3480b252009-01-14 14:14:16 +01002956SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002957{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002958 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002959}
2960
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961/*
2962 * The worst of all namespace operations - renaming directory. "Perverted"
2963 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2964 * Problems:
2965 * a) we can get into loop creation. Check is done in is_subdir().
2966 * b) race potential - two innocent renames can create a loop together.
2967 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002968 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 * story.
2970 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002971 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 * whether the target exists). Solution: try to be smart with locking
2973 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002974 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 * move will be locked. Thus we can rank directories by the tree
2976 * (ancestors first) and rank all non-directories after them.
2977 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002978 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 * HOWEVER, it relies on the assumption that any object with ->lookup()
2980 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2981 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07002982 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002983 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002985 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 * locking].
2987 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002988static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2989 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990{
2991 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07002992 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
2994 /*
2995 * If we are going to change the parent - check write permissions,
2996 * we'll need to flip '..'.
2997 */
2998 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002999 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 if (error)
3001 return error;
3002 }
3003
3004 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3005 if (error)
3006 return error;
3007
Al Virod83c49f2010-04-30 17:17:09 -04003008 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003009 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003010
3011 error = -EBUSY;
3012 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3013 goto out;
3014
Sage Weil3cebde22011-05-29 21:20:59 -07003015 if (target)
3016 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003017 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3018 if (error)
3019 goto out;
3020
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003022 target->i_flags |= S_DEAD;
3023 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 }
Sage Weil9055cba2011-05-24 13:06:12 -07003025out:
3026 if (target)
3027 mutex_unlock(&target->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003028 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003029 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3030 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 return error;
3032}
3033
Adrian Bunk75c96f82005-05-05 16:16:09 -07003034static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3035 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036{
Sage Weil51892bb2011-05-24 13:06:13 -07003037 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 int error;
3039
3040 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3041 if (error)
3042 return error;
3043
3044 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003046 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003047
3048 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003050 goto out;
3051
3052 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3053 if (error)
3054 goto out;
3055
3056 if (target)
3057 dont_mount(new_dentry);
3058 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3059 d_move(old_dentry, new_dentry);
3060out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003062 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 dput(new_dentry);
3064 return error;
3065}
3066
3067int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3068 struct inode *new_dir, struct dentry *new_dentry)
3069{
3070 int error;
3071 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003072 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073
3074 if (old_dentry->d_inode == new_dentry->d_inode)
3075 return 0;
3076
3077 error = may_delete(old_dir, old_dentry, is_dir);
3078 if (error)
3079 return error;
3080
3081 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003082 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 else
3084 error = may_delete(new_dir, new_dentry, is_dir);
3085 if (error)
3086 return error;
3087
Al Viroacfa4382008-12-04 10:06:33 -05003088 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 return -EPERM;
3090
Robert Love0eeca282005-07-12 17:06:03 -04003091 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3092
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 if (is_dir)
3094 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3095 else
3096 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003097 if (!error)
3098 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003099 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003100 fsnotify_oldname_free(old_name);
3101
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 return error;
3103}
3104
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003105SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3106 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107{
Al Viro2ad94ae2008-07-21 09:32:51 -04003108 struct dentry *old_dir, *new_dir;
3109 struct dentry *old_dentry, *new_dentry;
3110 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003112 char *from;
3113 char *to;
3114 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115
Al Viro2ad94ae2008-07-21 09:32:51 -04003116 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 if (error)
3118 goto exit;
3119
Al Viro2ad94ae2008-07-21 09:32:51 -04003120 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 if (error)
3122 goto exit1;
3123
3124 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003125 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 goto exit2;
3127
Jan Blunck4ac91372008-02-14 19:34:32 -08003128 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 error = -EBUSY;
3130 if (oldnd.last_type != LAST_NORM)
3131 goto exit2;
3132
Jan Blunck4ac91372008-02-14 19:34:32 -08003133 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 if (newnd.last_type != LAST_NORM)
3135 goto exit2;
3136
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003137 oldnd.flags &= ~LOOKUP_PARENT;
3138 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003139 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003140
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 trap = lock_rename(new_dir, old_dir);
3142
Christoph Hellwig49705b72005-11-08 21:35:06 -08003143 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 error = PTR_ERR(old_dentry);
3145 if (IS_ERR(old_dentry))
3146 goto exit3;
3147 /* source must exist */
3148 error = -ENOENT;
3149 if (!old_dentry->d_inode)
3150 goto exit4;
3151 /* unless the source is a directory trailing slashes give -ENOTDIR */
3152 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3153 error = -ENOTDIR;
3154 if (oldnd.last.name[oldnd.last.len])
3155 goto exit4;
3156 if (newnd.last.name[newnd.last.len])
3157 goto exit4;
3158 }
3159 /* source should not be ancestor of target */
3160 error = -EINVAL;
3161 if (old_dentry == trap)
3162 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003163 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 error = PTR_ERR(new_dentry);
3165 if (IS_ERR(new_dentry))
3166 goto exit4;
3167 /* target should not be an ancestor of source */
3168 error = -ENOTEMPTY;
3169 if (new_dentry == trap)
3170 goto exit5;
3171
Dave Hansen9079b1e2008-02-15 14:37:49 -08003172 error = mnt_want_write(oldnd.path.mnt);
3173 if (error)
3174 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003175 error = security_path_rename(&oldnd.path, old_dentry,
3176 &newnd.path, new_dentry);
3177 if (error)
3178 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 error = vfs_rename(old_dir->d_inode, old_dentry,
3180 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003181exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003182 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183exit5:
3184 dput(new_dentry);
3185exit4:
3186 dput(old_dentry);
3187exit3:
3188 unlock_rename(new_dir, old_dir);
3189exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003190 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003191 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003193 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003195exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 return error;
3197}
3198
Heiko Carstensa26eab22009-01-14 14:14:17 +01003199SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003200{
3201 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3202}
3203
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3205{
3206 int len;
3207
3208 len = PTR_ERR(link);
3209 if (IS_ERR(link))
3210 goto out;
3211
3212 len = strlen(link);
3213 if (len > (unsigned) buflen)
3214 len = buflen;
3215 if (copy_to_user(buffer, link, len))
3216 len = -EFAULT;
3217out:
3218 return len;
3219}
3220
3221/*
3222 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3223 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3224 * using) it for any given inode is up to filesystem.
3225 */
3226int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3227{
3228 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003229 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003230 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003231
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003233 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003234 if (IS_ERR(cookie))
3235 return PTR_ERR(cookie);
3236
3237 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3238 if (dentry->d_inode->i_op->put_link)
3239 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3240 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241}
3242
3243int vfs_follow_link(struct nameidata *nd, const char *link)
3244{
3245 return __vfs_follow_link(nd, link);
3246}
3247
3248/* get the link contents into pagecache */
3249static char *page_getlink(struct dentry * dentry, struct page **ppage)
3250{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003251 char *kaddr;
3252 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003254 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003256 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003258 kaddr = kmap(page);
3259 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3260 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261}
3262
3263int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3264{
3265 struct page *page = NULL;
3266 char *s = page_getlink(dentry, &page);
3267 int res = vfs_readlink(dentry,buffer,buflen,s);
3268 if (page) {
3269 kunmap(page);
3270 page_cache_release(page);
3271 }
3272 return res;
3273}
3274
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003275void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003277 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003279 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280}
3281
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003282void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003284 struct page *page = cookie;
3285
3286 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 kunmap(page);
3288 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 }
3290}
3291
Nick Piggin54566b22009-01-04 12:00:53 -08003292/*
3293 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3294 */
3295int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296{
3297 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003298 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003299 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003300 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003302 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3303 if (nofs)
3304 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305
NeilBrown7e53cac2006-03-25 03:07:57 -08003306retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003307 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003308 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003310 goto fail;
3311
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 kaddr = kmap_atomic(page, KM_USER0);
3313 memcpy(kaddr, symname, len-1);
3314 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003315
3316 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3317 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 if (err < 0)
3319 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003320 if (err < len-1)
3321 goto retry;
3322
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 mark_inode_dirty(inode);
3324 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325fail:
3326 return err;
3327}
3328
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003329int page_symlink(struct inode *inode, const char *symname, int len)
3330{
3331 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003332 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003333}
3334
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003335const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336 .readlink = generic_readlink,
3337 .follow_link = page_follow_link_light,
3338 .put_link = page_put_link,
3339};
3340
Al Viro2d8f3032008-07-22 09:59:21 -04003341EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003342EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343EXPORT_SYMBOL(follow_down);
3344EXPORT_SYMBOL(follow_up);
3345EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3346EXPORT_SYMBOL(getname);
3347EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348EXPORT_SYMBOL(lookup_one_len);
3349EXPORT_SYMBOL(page_follow_link_light);
3350EXPORT_SYMBOL(page_put_link);
3351EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003352EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353EXPORT_SYMBOL(page_symlink);
3354EXPORT_SYMBOL(page_symlink_inode_operations);
Al Viroc9c6cac2011-02-16 15:15:47 -05003355EXPORT_SYMBOL(kern_path_parent);
Al Virod1811462008-08-02 00:49:18 -04003356EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003357EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003358EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08003359EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360EXPORT_SYMBOL(unlock_rename);
3361EXPORT_SYMBOL(vfs_create);
3362EXPORT_SYMBOL(vfs_follow_link);
3363EXPORT_SYMBOL(vfs_link);
3364EXPORT_SYMBOL(vfs_mkdir);
3365EXPORT_SYMBOL(vfs_mknod);
3366EXPORT_SYMBOL(generic_permission);
3367EXPORT_SYMBOL(vfs_readlink);
3368EXPORT_SYMBOL(vfs_rename);
3369EXPORT_SYMBOL(vfs_rmdir);
3370EXPORT_SYMBOL(vfs_symlink);
3371EXPORT_SYMBOL(vfs_unlink);
3372EXPORT_SYMBOL(dentry_unhash);
3373EXPORT_SYMBOL(generic_readlink);