blob: 5362af9b73729905817a76b5c24a5b33c48d5ce7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
Eric Parise81e3f42009-12-04 15:47:36 -050037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800117static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400161 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700169/*
170 * This does basic POSIX ACL permission checking
171 */
172static int acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask))
174{
175 umode_t mode = inode->i_mode;
176
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
179 if (current_fsuid() == inode->i_uid)
180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask);
184 if (error != -EAGAIN)
185 return error;
186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
195 if ((mask & ~mode) == 0)
196 return 0;
197 return -EACCES;
198}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200/**
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 *
206 * Used to check for read/write/execute permissions on a file.
207 * We use "fsuid" for this, letting us set arbitrary permissions
208 * for filesystem access without changing the "normal" uids which
209 * are used for other things..
210 */
211int generic_permission(struct inode *inode, int mask,
212 int (*check_acl)(struct inode *inode, int mask))
213{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700214 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700217 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 */
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700219 ret = acl_permission_check(inode, mask, check_acl);
220 if (ret != -EACCES)
221 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 /*
224 * Read/write DACs are always overridable.
225 * Executable DACs are overridable if at least one exec bit is set.
226 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200227 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 if (capable(CAP_DAC_OVERRIDE))
229 return 0;
230
231 /*
232 * Searching includes executable on directories, else just read.
233 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600234 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
237 return 0;
238
239 return -EACCES;
240}
241
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200242/**
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
246 *
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
251 */
Al Virof419a2e2008-07-22 00:07:17 -0400252int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
Al Viroe6305c42008-07-15 21:03:57 -0400254 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700257 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 /*
260 * Nobody gets write access to a read-only fs.
261 */
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
264 return -EROFS;
265
266 /*
267 * Nobody gets write access to an immutable file.
268 */
269 if (IS_IMMUTABLE(inode))
270 return -EACCES;
271 }
272
Al Viroacfa4382008-12-04 10:06:33 -0500273 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400274 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200275 else
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (retval)
279 return retval;
280
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700281 retval = devcgroup_inode_permission(inode, mask);
282 if (retval)
283 return retval;
284
Eric Parisd09ca732010-07-23 11:43:57 -0400285 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
287
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800288/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800289 * file_permission - check for additional access rights to a given file
290 * @file: file to check access rights for
291 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
292 *
293 * Used to check for read/write/execute permissions on an already opened
294 * file.
295 *
296 * Note:
297 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200298 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800299 */
300int file_permission(struct file *file, int mask)
301{
Al Virof419a2e2008-07-22 00:07:17 -0400302 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800303}
304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305/*
306 * get_write_access() gets write permission for a file.
307 * put_write_access() releases this write permission.
308 * This is used for regular files.
309 * We cannot support write (and maybe mmap read-write shared) accesses and
310 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
311 * can have the following values:
312 * 0: no writers, no VM_DENYWRITE mappings
313 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
314 * > 0: (i_writecount) users are writing to the file.
315 *
316 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
317 * except for the cases where we don't hold i_writecount yet. Then we need to
318 * use {get,deny}_write_access() - these functions check the sign and refuse
319 * to do the change if sign is wrong. Exclusion between them is provided by
320 * the inode->i_lock spinlock.
321 */
322
323int get_write_access(struct inode * inode)
324{
325 spin_lock(&inode->i_lock);
326 if (atomic_read(&inode->i_writecount) < 0) {
327 spin_unlock(&inode->i_lock);
328 return -ETXTBSY;
329 }
330 atomic_inc(&inode->i_writecount);
331 spin_unlock(&inode->i_lock);
332
333 return 0;
334}
335
336int deny_write_access(struct file * file)
337{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800338 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 spin_lock(&inode->i_lock);
341 if (atomic_read(&inode->i_writecount) > 0) {
342 spin_unlock(&inode->i_lock);
343 return -ETXTBSY;
344 }
345 atomic_dec(&inode->i_writecount);
346 spin_unlock(&inode->i_lock);
347
348 return 0;
349}
350
Jan Blunck1d957f92008-02-14 19:34:35 -0800351/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800352 * path_get - get a reference to a path
353 * @path: path to get the reference to
354 *
355 * Given a path increment the reference count to the dentry and the vfsmount.
356 */
357void path_get(struct path *path)
358{
359 mntget(path->mnt);
360 dget(path->dentry);
361}
362EXPORT_SYMBOL(path_get);
363
364/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800365 * path_put - put a reference to a path
366 * @path: path to put the reference to
367 *
368 * Given a path decrement the reference count to the dentry and the vfsmount.
369 */
370void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Jan Blunck1d957f92008-02-14 19:34:35 -0800372 dput(path->dentry);
373 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374}
Jan Blunck1d957f92008-02-14 19:34:35 -0800375EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Trond Myklebust834f2a42005-10-18 14:20:16 -0700377/**
378 * release_open_intent - free up open intent resources
379 * @nd: pointer to nameidata
380 */
381void release_open_intent(struct nameidata *nd)
382{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800383 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700384 put_filp(nd->intent.open.file);
385 else
386 fput(nd->intent.open.file);
387}
388
Ian Kentbcdc5e02006-09-27 01:50:44 -0700389static inline struct dentry *
390do_revalidate(struct dentry *dentry, struct nameidata *nd)
391{
392 int status = dentry->d_op->d_revalidate(dentry, nd);
393 if (unlikely(status <= 0)) {
394 /*
395 * The dentry failed validation.
396 * If d_revalidate returned 0 attempt to invalidate
397 * the dentry otherwise d_revalidate is asking us
398 * to return a fail status.
399 */
400 if (!status) {
401 if (!d_invalidate(dentry)) {
402 dput(dentry);
403 dentry = NULL;
404 }
405 } else {
406 dput(dentry);
407 dentry = ERR_PTR(status);
408 }
409 }
410 return dentry;
411}
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413/*
Jeff Layton39159de2009-12-07 12:01:50 -0500414 * force_reval_path - force revalidation of a dentry
415 *
416 * In some situations the path walking code will trust dentries without
417 * revalidating them. This causes problems for filesystems that depend on
418 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
419 * (which indicates that it's possible for the dentry to go stale), force
420 * a d_revalidate call before proceeding.
421 *
422 * Returns 0 if the revalidation was successful. If the revalidation fails,
423 * either return the error returned by d_revalidate or -ESTALE if the
424 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
425 * invalidate the dentry. It's up to the caller to handle putting references
426 * to the path if necessary.
427 */
428static int
429force_reval_path(struct path *path, struct nameidata *nd)
430{
431 int status;
432 struct dentry *dentry = path->dentry;
433
434 /*
435 * only check on filesystems where it's possible for the dentry to
436 * become stale. It's assumed that if this flag is set then the
437 * d_revalidate op will also be defined.
438 */
439 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
440 return 0;
441
442 status = dentry->d_op->d_revalidate(dentry, nd);
443 if (status > 0)
444 return 0;
445
446 if (!status) {
447 d_invalidate(dentry);
448 status = -ESTALE;
449 }
450 return status;
451}
452
453/*
Al Virob75b5082009-12-16 01:01:38 -0500454 * Short-cut version of permission(), for calling on directories
455 * during pathname resolution. Combines parts of permission()
456 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 *
458 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500459 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 * complete permission check.
461 */
Al Virob75b5082009-12-16 01:01:38 -0500462static int exec_permission(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700464 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700466 if (inode->i_op->permission) {
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700467 ret = inode->i_op->permission(inode, MAY_EXEC);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700468 if (!ret)
469 goto ok;
470 return ret;
471 }
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700472 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
473 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 goto ok;
475
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700476 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 goto ok;
478
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700479 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480ok:
Al Virob77b0642008-07-17 09:37:02 -0400481 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Al Viro2a737872009-04-07 11:49:53 -0400484static __always_inline void set_root(struct nameidata *nd)
485{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200486 if (!nd->root.mnt)
487 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400488}
489
Al Viro6de88d72009-08-09 01:41:57 +0400490static int link_path_walk(const char *, struct nameidata *);
491
Arjan van de Venf1662352006-01-14 13:21:31 -0800492static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (IS_ERR(link))
495 goto fail;
496
497 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400498 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800499 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400500 nd->path = nd->root;
501 path_get(&nd->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100503
Al Virodef4af32009-12-26 08:37:05 -0500504 return link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800506 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 return PTR_ERR(link);
508}
509
Jan Blunck1d957f92008-02-14 19:34:35 -0800510static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700511{
512 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800513 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700514 mntput(path->mnt);
515}
516
517static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
518{
Jan Blunck4ac91372008-02-14 19:34:32 -0800519 dput(nd->path.dentry);
Huang Shijie9a229682010-04-02 17:37:13 +0800520 if (nd->path.mnt != path->mnt) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800521 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800522 nd->path.mnt = path->mnt;
523 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800524 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700525}
526
Al Virodef4af32009-12-26 08:37:05 -0500527static __always_inline int
528__do_follow_link(struct path *path, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800529{
530 int error;
Ian Kent051d3812006-03-27 01:14:53 -0800531 struct dentry *dentry = path->dentry;
532
533 touch_atime(path->mnt, dentry);
534 nd_set_link(nd, NULL);
535
Jan Blunck4ac91372008-02-14 19:34:32 -0800536 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800537 path_to_nameidata(path, nd);
538 dget(dentry);
539 }
540 mntget(path->mnt);
Al Viro86acdca12009-12-22 23:45:11 -0500541 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500542 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
543 error = PTR_ERR(*p);
544 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800545 char *s = nd_get_link(nd);
546 error = 0;
547 if (s)
548 error = __vfs_follow_link(nd, s);
Jeff Layton39159de2009-12-07 12:01:50 -0500549 else if (nd->last_type == LAST_BIND) {
550 error = force_reval_path(&nd->path, nd);
551 if (error)
552 path_put(&nd->path);
553 }
Ian Kent051d3812006-03-27 01:14:53 -0800554 }
Ian Kent051d3812006-03-27 01:14:53 -0800555 return error;
556}
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558/*
559 * This limits recursive symlink follows to 8, while
560 * limiting consecutive symlinks to 40.
561 *
562 * Without that kind of total limit, nasty chains of consecutive
563 * symlinks can cause almost arbitrarily long lookups.
564 */
Al Viro90ebe562005-06-06 13:35:58 -0700565static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
Al Virodef4af32009-12-26 08:37:05 -0500567 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 int err = -ELOOP;
569 if (current->link_count >= MAX_NESTED_LINKS)
570 goto loop;
571 if (current->total_link_count >= 40)
572 goto loop;
573 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
574 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700575 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (err)
577 goto loop;
578 current->link_count++;
579 current->total_link_count++;
580 nd->depth++;
Al Virodef4af32009-12-26 08:37:05 -0500581 err = __do_follow_link(path, nd, &cookie);
582 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
583 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
Al Viro258fa992009-08-09 01:32:02 +0400584 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 current->link_count--;
586 nd->depth--;
587 return err;
588loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800589 path_put_conditional(path, nd);
590 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 return err;
592}
593
Al Virobab77eb2009-04-18 03:26:48 -0400594int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595{
596 struct vfsmount *parent;
597 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000598
599 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400600 parent = path->mnt->mnt_parent;
601 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000602 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 return 0;
604 }
605 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400606 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000607 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400608 dput(path->dentry);
609 path->dentry = mountpoint;
610 mntput(path->mnt);
611 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 return 1;
613}
614
615/* no need for dcache_lock, as serialization is taken care in
616 * namespace.c
617 */
Al Viro463ffb22005-06-06 13:36:05 -0700618static int __follow_mount(struct path *path)
619{
620 int res = 0;
621 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400622 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700623 if (!mounted)
624 break;
625 dput(path->dentry);
626 if (res)
627 mntput(path->mnt);
628 path->mnt = mounted;
629 path->dentry = dget(mounted->mnt_root);
630 res = 1;
631 }
632 return res;
633}
634
Al Viro79ed0222009-04-18 13:59:41 -0400635static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Al Viro79ed0222009-04-18 13:59:41 -0400637 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400638 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (!mounted)
640 break;
Al Viro79ed0222009-04-18 13:59:41 -0400641 dput(path->dentry);
642 mntput(path->mnt);
643 path->mnt = mounted;
644 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
648/* no need for dcache_lock, as serialization is taken care in
649 * namespace.c
650 */
Al Viro9393bd02009-04-18 13:58:15 -0400651int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653 struct vfsmount *mounted;
654
Al Viro1c755af2009-04-18 14:06:57 -0400655 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400657 dput(path->dentry);
658 mntput(path->mnt);
659 path->mnt = mounted;
660 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 return 1;
662 }
663 return 0;
664}
665
Arjan van de Venf1662352006-01-14 13:21:31 -0800666static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Al Viro2a737872009-04-07 11:49:53 -0400668 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800671 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Al Viro2a737872009-04-07 11:49:53 -0400673 if (nd->path.dentry == nd->root.dentry &&
674 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 break;
676 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800677 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -0500678 /* rare case of legitimate dget_parent()... */
679 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 dput(old);
681 break;
682 }
Al Viro3088dd72010-01-30 15:47:29 -0500683 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 }
Al Viro79ed0222009-04-18 13:59:41 -0400686 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689/*
Nick Pigginbaa03892010-08-18 04:37:31 +1000690 * Allocate a dentry with name and parent, and perform a parent
691 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
692 * on error. parent->d_inode->i_mutex must be held. d_lookup must
693 * have verified that no child exists while under i_mutex.
694 */
695static struct dentry *d_alloc_and_lookup(struct dentry *parent,
696 struct qstr *name, struct nameidata *nd)
697{
698 struct inode *inode = parent->d_inode;
699 struct dentry *dentry;
700 struct dentry *old;
701
702 /* Don't create child dentry for a dead directory. */
703 if (unlikely(IS_DEADDIR(inode)))
704 return ERR_PTR(-ENOENT);
705
706 dentry = d_alloc(parent, name);
707 if (unlikely(!dentry))
708 return ERR_PTR(-ENOMEM);
709
710 old = inode->i_op->lookup(inode, dentry, nd);
711 if (unlikely(old)) {
712 dput(dentry);
713 dentry = old;
714 }
715 return dentry;
716}
717
718/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 * It's more convoluted than I'd like it to be, but... it's still fairly
720 * small and for now I'd prefer to have fast path as straight as possible.
721 * It _is_ time-critical.
722 */
723static int do_lookup(struct nameidata *nd, struct qstr *name,
724 struct path *path)
725{
Jan Blunck4ac91372008-02-14 19:34:32 -0800726 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400727 struct dentry *dentry, *parent;
728 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400729 /*
730 * See if the low-level filesystem might want
731 * to use its own hash..
732 */
733 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
734 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
735 if (err < 0)
736 return err;
737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Nick Pigginb04f7842010-08-18 04:37:34 +1000739 /*
740 * Rename seqlock is not required here because in the off chance
741 * of a false negative due to a concurrent rename, we're going to
742 * do the non-racy lookup, below.
743 */
Al Viro3cac2602009-08-13 18:27:43 +0400744 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (!dentry)
746 goto need_lookup;
Nick Piggin2e2e88e2010-08-18 04:37:30 +1000747found:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 if (dentry->d_op && dentry->d_op->d_revalidate)
749 goto need_revalidate;
750done:
751 path->mnt = mnt;
752 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700753 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return 0;
755
756need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400757 parent = nd->path.dentry;
758 dir = parent->d_inode;
759
760 mutex_lock(&dir->i_mutex);
761 /*
762 * First re-do the cached lookup just in case it was created
Nick Pigginb04f7842010-08-18 04:37:34 +1000763 * while we waited for the directory semaphore, or the first
764 * lookup failed due to an unrelated rename.
Al Viro6e6b1bd2009-08-13 23:38:37 +0400765 *
Nick Pigginb04f7842010-08-18 04:37:34 +1000766 * This could use version numbering or similar to avoid unnecessary
767 * cache lookups, but then we'd have to do the first lookup in the
768 * non-racy way. However in the common case here, everything should
769 * be hot in cache, so would it be a big win?
Al Viro6e6b1bd2009-08-13 23:38:37 +0400770 */
771 dentry = d_lookup(parent, name);
Nick Pigginbaa03892010-08-18 04:37:31 +1000772 if (likely(!dentry)) {
773 dentry = d_alloc_and_lookup(parent, name, nd);
Al Viro6e6b1bd2009-08-13 23:38:37 +0400774 mutex_unlock(&dir->i_mutex);
775 if (IS_ERR(dentry))
776 goto fail;
777 goto done;
778 }
Al Viro6e6b1bd2009-08-13 23:38:37 +0400779 /*
780 * Uhhuh! Nasty case: the cache was re-populated while
781 * we waited on the semaphore. Need to revalidate.
782 */
783 mutex_unlock(&dir->i_mutex);
Nick Piggin2e2e88e2010-08-18 04:37:30 +1000784 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
786need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700787 dentry = do_revalidate(dentry, nd);
788 if (!dentry)
789 goto need_lookup;
790 if (IS_ERR(dentry))
791 goto fail;
792 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794fail:
795 return PTR_ERR(dentry);
796}
797
798/*
Al Viroac278a92010-02-16 18:09:36 +0000799 * This is a temporary kludge to deal with "automount" symlinks; proper
800 * solution is to trigger them on follow_mount(), so that do_lookup()
801 * would DTRT. To be killed before 2.6.34-final.
802 */
803static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
804{
805 return inode && unlikely(inode->i_op->follow_link) &&
806 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
807}
808
809/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100811 * This is the basic name resolution function, turning a pathname into
812 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100814 * Returns 0 and nd will have valid dentry and mnt on success.
815 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 */
Al Viro6de88d72009-08-09 01:41:57 +0400817static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
819 struct path next;
820 struct inode *inode;
821 int err;
822 unsigned int lookup_flags = nd->flags;
823
824 while (*name=='/')
825 name++;
826 if (!*name)
827 goto return_reval;
828
Jan Blunck4ac91372008-02-14 19:34:32 -0800829 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800831 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
833 /* At this point we know we have a real path component. */
834 for(;;) {
835 unsigned long hash;
836 struct qstr this;
837 unsigned int c;
838
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700839 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500840 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 if (err)
842 break;
843
844 this.name = name;
845 c = *(const unsigned char *)name;
846
847 hash = init_name_hash();
848 do {
849 name++;
850 hash = partial_name_hash(c, hash);
851 c = *(const unsigned char *)name;
852 } while (c && (c != '/'));
853 this.len = name - (const char *) this.name;
854 this.hash = end_name_hash(hash);
855
856 /* remove trailing slashes? */
857 if (!c)
858 goto last_component;
859 while (*++name == '/');
860 if (!*name)
861 goto last_with_slashes;
862
863 /*
864 * "." and ".." are special - ".." especially so because it has
865 * to be able to know about the current root directory and
866 * parent relationships.
867 */
868 if (this.name[0] == '.') switch (this.len) {
869 default:
870 break;
871 case 2:
872 if (this.name[1] != '.')
873 break;
Al Viro58c465e2005-06-06 13:36:13 -0700874 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800875 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 /* fallthrough */
877 case 1:
878 continue;
879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 /* This does the actual lookups.. */
881 err = do_lookup(nd, &this, &next);
882 if (err)
883 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 err = -ENOENT;
886 inode = next.dentry->d_inode;
887 if (!inode)
888 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700891 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 if (err)
893 goto return_err;
894 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800895 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (!inode)
897 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700898 } else
899 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 err = -ENOTDIR;
901 if (!inode->i_op->lookup)
902 break;
903 continue;
904 /* here ends the main loop */
905
906last_with_slashes:
907 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
908last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800909 /* Clear LOOKUP_CONTINUE iff it was previously unset */
910 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (lookup_flags & LOOKUP_PARENT)
912 goto lookup_parent;
913 if (this.name[0] == '.') switch (this.len) {
914 default:
915 break;
916 case 2:
917 if (this.name[1] != '.')
918 break;
Al Viro58c465e2005-06-06 13:36:13 -0700919 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800920 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 /* fallthrough */
922 case 1:
923 goto return_reval;
924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 err = do_lookup(nd, &this, &next);
926 if (err)
927 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 inode = next.dentry->d_inode;
Al Viroac278a92010-02-16 18:09:36 +0000929 if (follow_on_final(inode, lookup_flags)) {
Al Viro90ebe562005-06-06 13:35:58 -0700930 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 if (err)
932 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800933 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700934 } else
935 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 err = -ENOENT;
937 if (!inode)
938 break;
939 if (lookup_flags & LOOKUP_DIRECTORY) {
940 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500941 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 break;
943 }
944 goto return_base;
945lookup_parent:
946 nd->last = this;
947 nd->last_type = LAST_NORM;
948 if (this.name[0] != '.')
949 goto return_base;
950 if (this.len == 1)
951 nd->last_type = LAST_DOT;
952 else if (this.len == 2 && this.name[1] == '.')
953 nd->last_type = LAST_DOTDOT;
954 else
955 goto return_base;
956return_reval:
957 /*
958 * We bypassed the ordinary revalidation routines.
959 * We may need to check the cached dentry for staleness.
960 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800961 if (nd->path.dentry && nd->path.dentry->d_sb &&
962 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 err = -ESTALE;
964 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800965 if (!nd->path.dentry->d_op->d_revalidate(
966 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 break;
968 }
969return_base:
970 return 0;
971out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800972 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 break;
974 }
Jan Blunck1d957f92008-02-14 19:34:35 -0800975 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976return_err:
977 return err;
978}
979
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800980static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
Al Viro6de88d72009-08-09 01:41:57 +0400982 struct path save = nd->path;
983 int result;
984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +0400986
987 /* make sure the stuff we saved doesn't go away */
988 path_get(&save);
989
990 result = link_path_walk(name, nd);
991 if (result == -ESTALE) {
992 /* nd->path had been dropped */
993 current->total_link_count = 0;
994 nd->path = save;
995 path_get(&nd->path);
996 nd->flags |= LOOKUP_REVAL;
997 result = link_path_walk(name, nd);
998 }
999
1000 path_put(&save);
1001
1002 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
1004
Al Viro9b4a9b12009-04-07 11:44:16 -04001005static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001007 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001008 int fput_needed;
1009 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1012 nd->flags = flags;
1013 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001014 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001017 set_root(nd);
1018 nd->path = nd->root;
1019 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001020 } else if (dfd == AT_FDCWD) {
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02001021 get_fs_pwd(current->fs, &nd->path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001022 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001023 struct dentry *dentry;
1024
1025 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001026 retval = -EBADF;
1027 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001028 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001029
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001030 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001031
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001032 retval = -ENOTDIR;
1033 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001034 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001035
1036 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001037 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001038 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001039
Jan Blunck5dd784d2008-02-14 19:34:38 -08001040 nd->path = file->f_path;
1041 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001042
1043 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001045 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001046
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001047fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001048 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001049out_fail:
1050 return retval;
1051}
1052
1053/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1054static int do_path_lookup(int dfd, const char *name,
1055 unsigned int flags, struct nameidata *nd)
1056{
1057 int retval = path_init(dfd, name, flags, nd);
1058 if (!retval)
1059 retval = path_walk(name, nd);
1060 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1061 nd->path.dentry->d_inode))
1062 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001063 if (nd->root.mnt) {
1064 path_put(&nd->root);
1065 nd->root.mnt = NULL;
1066 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001067 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001070int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001071 struct nameidata *nd)
1072{
1073 return do_path_lookup(AT_FDCWD, name, flags, nd);
1074}
1075
Al Virod1811462008-08-02 00:49:18 -04001076int kern_path(const char *name, unsigned int flags, struct path *path)
1077{
1078 struct nameidata nd;
1079 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1080 if (!res)
1081 *path = nd.path;
1082 return res;
1083}
1084
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001085/**
1086 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1087 * @dentry: pointer to dentry of the base directory
1088 * @mnt: pointer to vfs mount of the base directory
1089 * @name: pointer to file name
1090 * @flags: lookup flags
1091 * @nd: pointer to nameidata
1092 */
1093int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1094 const char *name, unsigned int flags,
1095 struct nameidata *nd)
1096{
1097 int retval;
1098
1099 /* same as do_path_lookup */
1100 nd->last_type = LAST_ROOT;
1101 nd->flags = flags;
1102 nd->depth = 0;
1103
Jan Blunckc8e7f442008-06-09 16:40:35 -07001104 nd->path.dentry = dentry;
1105 nd->path.mnt = mnt;
1106 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001107 nd->root = nd->path;
1108 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001109
1110 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001111 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1112 nd->path.dentry->d_inode))
1113 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001114
Al Viro5b857112009-04-07 11:53:49 -04001115 path_put(&nd->root);
1116 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001117
Al Viro2a737872009-04-07 11:49:53 -04001118 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001119}
1120
Christoph Hellwigeead1912007-10-16 23:25:38 -07001121static struct dentry *__lookup_hash(struct qstr *name,
1122 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001124 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001125 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 int err;
1127
Christoph Hellwig81fca442010-10-06 10:47:47 +02001128 err = exec_permission(inode);
1129 if (err)
1130 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132 /*
1133 * See if the low-level filesystem might want
1134 * to use its own hash..
1135 */
1136 if (base->d_op && base->d_op->d_hash) {
1137 err = base->d_op->d_hash(base, name);
1138 dentry = ERR_PTR(err);
1139 if (err < 0)
1140 goto out;
1141 }
1142
Nick Pigginb04f7842010-08-18 04:37:34 +10001143 /*
1144 * Don't bother with __d_lookup: callers are for creat as
1145 * well as unlink, so a lot of the time it would cost
1146 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001147 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001148 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001149
1150 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1151 dentry = do_revalidate(dentry, nd);
1152
Nick Pigginbaa03892010-08-18 04:37:31 +10001153 if (!dentry)
1154 dentry = d_alloc_and_lookup(base, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155out:
1156 return dentry;
1157}
1158
James Morris057f6c02007-04-26 00:12:05 -07001159/*
1160 * Restricted form of lookup. Doesn't follow links, single-component only,
1161 * needs parent already locked. Doesn't follow mounts.
1162 * SMP-safe.
1163 */
Adrian Bunka244e162006-03-31 02:32:11 -08001164static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165{
Jan Blunck4ac91372008-02-14 19:34:32 -08001166 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}
1168
Christoph Hellwigeead1912007-10-16 23:25:38 -07001169static int __lookup_one_len(const char *name, struct qstr *this,
1170 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171{
1172 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 unsigned int c;
1174
James Morris057f6c02007-04-26 00:12:05 -07001175 this->name = name;
1176 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001178 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 hash = init_name_hash();
1181 while (len--) {
1182 c = *(const unsigned char *)name++;
1183 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001184 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 hash = partial_name_hash(c, hash);
1186 }
James Morris057f6c02007-04-26 00:12:05 -07001187 this->hash = end_name_hash(hash);
1188 return 0;
1189}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Christoph Hellwigeead1912007-10-16 23:25:38 -07001191/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001192 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001193 * @name: pathname component to lookup
1194 * @base: base directory to lookup from
1195 * @len: maximum length @len should be interpreted to
1196 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001197 * Note that this routine is purely a helper for filesystem usage and should
1198 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001199 * nameidata argument is passed to the filesystem methods and a filesystem
1200 * using this helper needs to be prepared for that.
1201 */
James Morris057f6c02007-04-26 00:12:05 -07001202struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1203{
1204 int err;
1205 struct qstr this;
1206
David Woodhouse2f9092e2009-04-20 23:18:37 +01001207 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1208
James Morris057f6c02007-04-26 00:12:05 -07001209 err = __lookup_one_len(name, &this, base, len);
1210 if (err)
1211 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001212
Christoph Hellwig49705b72005-11-08 21:35:06 -08001213 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001214}
1215
Al Viro2d8f3032008-07-22 09:59:21 -04001216int user_path_at(int dfd, const char __user *name, unsigned flags,
1217 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
Al Viro2d8f3032008-07-22 09:59:21 -04001219 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 char *tmp = getname(name);
1221 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001223
1224 BUG_ON(flags & LOOKUP_PARENT);
1225
1226 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001228 if (!err)
1229 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
1231 return err;
1232}
1233
Al Viro2ad94ae2008-07-21 09:32:51 -04001234static int user_path_parent(int dfd, const char __user *path,
1235 struct nameidata *nd, char **name)
1236{
1237 char *s = getname(path);
1238 int error;
1239
1240 if (IS_ERR(s))
1241 return PTR_ERR(s);
1242
1243 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1244 if (error)
1245 putname(s);
1246 else
1247 *name = s;
1248
1249 return error;
1250}
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252/*
1253 * It's inline, so penalty for filesystems that don't use sticky bit is
1254 * minimal.
1255 */
1256static inline int check_sticky(struct inode *dir, struct inode *inode)
1257{
David Howellsda9592e2008-11-14 10:39:05 +11001258 uid_t fsuid = current_fsuid();
1259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 if (!(dir->i_mode & S_ISVTX))
1261 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001262 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001264 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 return 0;
1266 return !capable(CAP_FOWNER);
1267}
1268
1269/*
1270 * Check whether we can remove a link victim from directory dir, check
1271 * whether the type of victim is right.
1272 * 1. We can't do it if dir is read-only (done in permission())
1273 * 2. We should have write and exec permissions on dir
1274 * 3. We can't remove anything from append-only dir
1275 * 4. We can't do anything with immutable dir (done in permission())
1276 * 5. If the sticky bit on dir is set we should either
1277 * a. be owner of dir, or
1278 * b. be owner of victim, or
1279 * c. have CAP_FOWNER capability
1280 * 6. If the victim is append-only or immutable we can't do antyhing with
1281 * links pointing to it.
1282 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1283 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1284 * 9. We can't remove a root or mountpoint.
1285 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1286 * nfs_async_unlink().
1287 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001288static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
1290 int error;
1291
1292 if (!victim->d_inode)
1293 return -ENOENT;
1294
1295 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001296 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Al Virof419a2e2008-07-22 00:07:17 -04001298 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 if (error)
1300 return error;
1301 if (IS_APPEND(dir))
1302 return -EPERM;
1303 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001304 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 return -EPERM;
1306 if (isdir) {
1307 if (!S_ISDIR(victim->d_inode->i_mode))
1308 return -ENOTDIR;
1309 if (IS_ROOT(victim))
1310 return -EBUSY;
1311 } else if (S_ISDIR(victim->d_inode->i_mode))
1312 return -EISDIR;
1313 if (IS_DEADDIR(dir))
1314 return -ENOENT;
1315 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1316 return -EBUSY;
1317 return 0;
1318}
1319
1320/* Check whether we can create an object with dentry child in directory
1321 * dir.
1322 * 1. We can't do it if child already exists (open has special treatment for
1323 * this case, but since we are inlined it's OK)
1324 * 2. We can't do it if dir is read-only (done in permission())
1325 * 3. We should have write and exec permissions on dir
1326 * 4. We can't do it if dir is immutable (done in permission())
1327 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001328static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
1330 if (child->d_inode)
1331 return -EEXIST;
1332 if (IS_DEADDIR(dir))
1333 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001334 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335}
1336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337/*
1338 * p1 and p2 should be directories on the same fs.
1339 */
1340struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1341{
1342 struct dentry *p;
1343
1344 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001345 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 return NULL;
1347 }
1348
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001349 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001351 p = d_ancestor(p2, p1);
1352 if (p) {
1353 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1354 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1355 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001358 p = d_ancestor(p1, p2);
1359 if (p) {
1360 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1361 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1362 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364
Ingo Molnarf2eace22006-07-03 00:25:05 -07001365 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1366 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 return NULL;
1368}
1369
1370void unlock_rename(struct dentry *p1, struct dentry *p2)
1371{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001372 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001374 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001375 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
1377}
1378
1379int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1380 struct nameidata *nd)
1381{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001382 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 if (error)
1385 return error;
1386
Al Viroacfa4382008-12-04 10:06:33 -05001387 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 return -EACCES; /* shouldn't it be ENOSYS? */
1389 mode &= S_IALLUGO;
1390 mode |= S_IFREG;
1391 error = security_inode_create(dir, dentry, mode);
1392 if (error)
1393 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001395 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001396 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 return error;
1398}
1399
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001400int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001402 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 struct inode *inode = dentry->d_inode;
1404 int error;
1405
1406 if (!inode)
1407 return -ENOENT;
1408
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001409 switch (inode->i_mode & S_IFMT) {
1410 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001412 case S_IFDIR:
1413 if (acc_mode & MAY_WRITE)
1414 return -EISDIR;
1415 break;
1416 case S_IFBLK:
1417 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001418 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001420 /*FALLTHRU*/
1421 case S_IFIFO:
1422 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001424 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001425 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001426
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001427 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001428 if (error)
1429 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001430
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 /*
1432 * An append-only file must be opened in append mode for writing.
1433 */
1434 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001435 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001436 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001438 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
1440
1441 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001442 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1443 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
1445 /*
1446 * Ensure there are no outstanding leases on the file.
1447 */
Al Virob65a9cf2009-12-16 06:27:40 -05001448 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001449}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
Al Viro7715b522009-12-16 03:54:00 -05001451static int handle_truncate(struct path *path)
1452{
1453 struct inode *inode = path->dentry->d_inode;
1454 int error = get_write_access(inode);
1455 if (error)
1456 return error;
1457 /*
1458 * Refuse to truncate files with mandatory locks held on them.
1459 */
1460 error = locks_verify_locked(inode);
1461 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09001462 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05001463 if (!error) {
1464 error = do_truncate(path->dentry, 0,
1465 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1466 NULL);
1467 }
1468 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001469 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
Dave Hansend57999e2008-02-15 14:37:27 -08001472/*
1473 * Be careful about ever adding any more callers of this
1474 * function. Its flags must be in the namei format, not
1475 * what get passed to sys_open().
1476 */
1477static int __open_namei_create(struct nameidata *nd, struct path *path,
Al Viro8737c932009-12-24 06:47:55 -05001478 int open_flag, int mode)
Dave Hansenaab520e2006-09-30 23:29:02 -07001479{
1480 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001481 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001482
1483 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001484 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001485 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1486 if (error)
1487 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001488 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001489out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001490 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001491 dput(nd->path.dentry);
1492 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001493 if (error)
1494 return error;
1495 /* Don't check for write permission, don't truncate */
Al Viro8737c932009-12-24 06:47:55 -05001496 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001497}
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499/*
Dave Hansend57999e2008-02-15 14:37:27 -08001500 * Note that while the flag value (low two bits) for sys_open means:
1501 * 00 - read-only
1502 * 01 - write-only
1503 * 10 - read-write
1504 * 11 - special
1505 * it is changed into
1506 * 00 - no permissions needed
1507 * 01 - read-permission
1508 * 10 - write-permission
1509 * 11 - read-write
1510 * for the internal routines (ie open_namei()/follow_link() etc)
1511 * This is more logical, and also allows the 00 "no perm needed"
1512 * to be used for symlinks (where the permissions are checked
1513 * later).
1514 *
1515*/
1516static inline int open_to_namei_flags(int flag)
1517{
1518 if ((flag+1) & O_ACCMODE)
1519 flag++;
1520 return flag;
1521}
1522
Al Viro7715b522009-12-16 03:54:00 -05001523static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001524{
1525 /*
1526 * We'll never write to the fs underlying
1527 * a device file.
1528 */
1529 if (special_file(inode->i_mode))
1530 return 0;
1531 return (flag & O_TRUNC);
1532}
1533
Al Viro648fa862009-12-24 01:26:48 -05001534static struct file *finish_open(struct nameidata *nd,
Al Viro9a661792009-12-24 06:49:47 -05001535 int open_flag, int acc_mode)
Al Viro648fa862009-12-24 01:26:48 -05001536{
1537 struct file *filp;
1538 int will_truncate;
1539 int error;
1540
Al Viro9a661792009-12-24 06:49:47 -05001541 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
Al Viro648fa862009-12-24 01:26:48 -05001542 if (will_truncate) {
1543 error = mnt_want_write(nd->path.mnt);
1544 if (error)
1545 goto exit;
1546 }
1547 error = may_open(&nd->path, acc_mode, open_flag);
1548 if (error) {
1549 if (will_truncate)
1550 mnt_drop_write(nd->path.mnt);
1551 goto exit;
1552 }
1553 filp = nameidata_to_filp(nd);
1554 if (!IS_ERR(filp)) {
1555 error = ima_file_check(filp, acc_mode);
1556 if (error) {
1557 fput(filp);
1558 filp = ERR_PTR(error);
1559 }
1560 }
1561 if (!IS_ERR(filp)) {
Al Viro648fa862009-12-24 01:26:48 -05001562 if (will_truncate) {
1563 error = handle_truncate(&nd->path);
1564 if (error) {
1565 fput(filp);
1566 filp = ERR_PTR(error);
1567 }
1568 }
1569 }
1570 /*
1571 * It is now safe to drop the mnt write
1572 * because the filp has had a write taken
1573 * on its behalf.
1574 */
1575 if (will_truncate)
1576 mnt_drop_write(nd->path.mnt);
Al Virod893f1b2010-10-29 03:30:42 -04001577 path_put(&nd->path);
Al Viro648fa862009-12-24 01:26:48 -05001578 return filp;
1579
1580exit:
1581 if (!IS_ERR(nd->intent.open.file))
1582 release_open_intent(nd);
1583 path_put(&nd->path);
1584 return ERR_PTR(error);
1585}
1586
Al Virofb1cc552009-12-24 01:58:28 -05001587static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viro5b369df2009-12-24 06:51:13 -05001588 int open_flag, int acc_mode,
Al Viro3e297b62010-03-26 12:40:13 -04001589 int mode, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05001590{
Al Viroa1e28032009-12-24 02:12:06 -05001591 struct dentry *dir = nd->path.dentry;
Al Virofb1cc552009-12-24 01:58:28 -05001592 struct file *filp;
Al Viro1f36f772009-12-26 10:56:19 -05001593 int error = -EISDIR;
Al Virofb1cc552009-12-24 01:58:28 -05001594
Al Viro1f36f772009-12-26 10:56:19 -05001595 switch (nd->last_type) {
1596 case LAST_DOTDOT:
1597 follow_dotdot(nd);
1598 dir = nd->path.dentry;
Neil Brown176306f2010-05-24 16:57:56 +10001599 case LAST_DOT:
Al Viro1f36f772009-12-26 10:56:19 -05001600 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1601 if (!dir->d_op->d_revalidate(dir, nd)) {
1602 error = -ESTALE;
1603 goto exit;
1604 }
1605 }
1606 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05001607 case LAST_ROOT:
1608 if (open_flag & O_CREAT)
1609 goto exit;
1610 /* fallthrough */
1611 case LAST_BIND:
1612 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05001613 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05001614 }
Al Viro67ee3ad2009-12-26 07:01:01 -05001615
Al Viro1f36f772009-12-26 10:56:19 -05001616 /* trailing slashes? */
1617 if (nd->last.name[nd->last.len]) {
1618 if (open_flag & O_CREAT)
1619 goto exit;
Jan Kara002baee2010-05-13 12:52:57 +02001620 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
Al Viro1f36f772009-12-26 10:56:19 -05001621 }
Al Viroa2c36b42009-12-24 03:39:50 -05001622
Al Viro1f36f772009-12-26 10:56:19 -05001623 /* just plain open? */
1624 if (!(open_flag & O_CREAT)) {
1625 error = do_lookup(nd, &nd->last, path);
1626 if (error)
1627 goto exit;
1628 error = -ENOENT;
1629 if (!path->dentry->d_inode)
1630 goto exit_dput;
1631 if (path->dentry->d_inode->i_op->follow_link)
1632 return NULL;
1633 error = -ENOTDIR;
Al Viro3e297b62010-03-26 12:40:13 -04001634 if (nd->flags & LOOKUP_DIRECTORY) {
1635 if (!path->dentry->d_inode->i_op->lookup)
1636 goto exit_dput;
1637 }
Al Viro1f36f772009-12-26 10:56:19 -05001638 path_to_nameidata(path, nd);
1639 audit_inode(pathname, nd->path.dentry);
1640 goto ok;
1641 }
1642
1643 /* OK, it's O_CREAT */
Al Viroa1e28032009-12-24 02:12:06 -05001644 mutex_lock(&dir->d_inode->i_mutex);
1645
1646 path->dentry = lookup_hash(nd);
1647 path->mnt = nd->path.mnt;
1648
Al Virofb1cc552009-12-24 01:58:28 -05001649 error = PTR_ERR(path->dentry);
1650 if (IS_ERR(path->dentry)) {
1651 mutex_unlock(&dir->d_inode->i_mutex);
1652 goto exit;
1653 }
1654
1655 if (IS_ERR(nd->intent.open.file)) {
1656 error = PTR_ERR(nd->intent.open.file);
1657 goto exit_mutex_unlock;
1658 }
1659
1660 /* Negative dentry, just create the file */
1661 if (!path->dentry->d_inode) {
1662 /*
1663 * This write is needed to ensure that a
1664 * ro->rw transition does not occur between
1665 * the time when the file is created and when
1666 * a permanent write count is taken through
1667 * the 'struct file' in nameidata_to_filp().
1668 */
1669 error = mnt_want_write(nd->path.mnt);
1670 if (error)
1671 goto exit_mutex_unlock;
1672 error = __open_namei_create(nd, path, open_flag, mode);
1673 if (error) {
1674 mnt_drop_write(nd->path.mnt);
1675 goto exit;
1676 }
1677 filp = nameidata_to_filp(nd);
1678 mnt_drop_write(nd->path.mnt);
Al Virod893f1b2010-10-29 03:30:42 -04001679 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05001680 if (!IS_ERR(filp)) {
1681 error = ima_file_check(filp, acc_mode);
1682 if (error) {
1683 fput(filp);
1684 filp = ERR_PTR(error);
1685 }
1686 }
1687 return filp;
1688 }
1689
1690 /*
1691 * It already exists.
1692 */
1693 mutex_unlock(&dir->d_inode->i_mutex);
1694 audit_inode(pathname, path->dentry);
1695
1696 error = -EEXIST;
Al Viro5b369df2009-12-24 06:51:13 -05001697 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05001698 goto exit_dput;
1699
1700 if (__follow_mount(path)) {
1701 error = -ELOOP;
Al Viro5b369df2009-12-24 06:51:13 -05001702 if (open_flag & O_NOFOLLOW)
Al Virofb1cc552009-12-24 01:58:28 -05001703 goto exit_dput;
1704 }
1705
1706 error = -ENOENT;
1707 if (!path->dentry->d_inode)
1708 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05001709
1710 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05001711 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05001712
1713 path_to_nameidata(path, nd);
1714 error = -EISDIR;
1715 if (S_ISDIR(path->dentry->d_inode->i_mode))
1716 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05001717ok:
Al Viro9a661792009-12-24 06:49:47 -05001718 filp = finish_open(nd, open_flag, acc_mode);
Al Virofb1cc552009-12-24 01:58:28 -05001719 return filp;
1720
1721exit_mutex_unlock:
1722 mutex_unlock(&dir->d_inode->i_mutex);
1723exit_dput:
1724 path_put_conditional(path, nd);
1725exit:
1726 if (!IS_ERR(nd->intent.open.file))
1727 release_open_intent(nd);
Al Virofb1cc552009-12-24 01:58:28 -05001728 path_put(&nd->path);
1729 return ERR_PTR(error);
1730}
1731
Dave Hansend57999e2008-02-15 14:37:27 -08001732/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001733 * Note that the low bits of the passed in "open_flag"
1734 * are not the same as in the local variable "flag". See
1735 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001737struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001738 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001740 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001741 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001742 int error;
Al Viro9850c052010-01-13 15:01:15 -05001743 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 int count = 0;
Dave Hansend57999e2008-02-15 14:37:27 -08001745 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001746 int force_reval = 0;
Al Viro1f36f772009-12-26 10:56:19 -05001747
1748 if (!(open_flag & O_CREAT))
1749 mode = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001751 /*
1752 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1753 * check for O_DSYNC if the need any syncing at all we enforce it's
1754 * always set instead of having to deal with possibly weird behaviour
1755 * for malicious applications setting only __O_SYNC.
1756 */
1757 if (open_flag & __O_SYNC)
1758 open_flag |= O_DSYNC;
1759
Al Viro6e8341a2009-04-06 11:16:22 -04001760 if (!acc_mode)
Al Viro6d125522009-12-24 06:58:56 -05001761 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
Trond Myklebust834f2a42005-10-18 14:20:16 -07001763 /* O_TRUNC implies we need access checks for write permissions */
Al Viro4296e2c2009-12-24 07:15:41 -05001764 if (open_flag & O_TRUNC)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001765 acc_mode |= MAY_WRITE;
1766
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 /* Allow the LSM permission hook to distinguish append
1768 access from general write access. */
Al Viro4296e2c2009-12-24 07:15:41 -05001769 if (open_flag & O_APPEND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 acc_mode |= MAY_APPEND;
1771
Al Viro1f36f772009-12-26 10:56:19 -05001772 /* find the parent */
Al Viro9850c052010-01-13 15:01:15 -05001773reval:
Al Viro9b4a9b12009-04-07 11:44:16 -04001774 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001776 return ERR_PTR(error);
Al Viro9850c052010-01-13 15:01:15 -05001777 if (force_reval)
1778 nd.flags |= LOOKUP_REVAL;
Al Viro38662482009-12-26 07:21:48 -05001779
1780 current->total_link_count = 0;
1781 error = link_path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001782 if (error) {
Al Viro10fa8e62009-12-26 07:09:49 -05001783 filp = ERR_PTR(error);
1784 goto out;
J. R. Okajima654f5622009-06-18 23:30:15 +09001785 }
Al Viro1f36f772009-12-26 10:56:19 -05001786 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
Al Viro9b4a9b12009-04-07 11:44:16 -04001787 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
1789 /*
Al Viroa2c36b42009-12-24 03:39:50 -05001790 * We have the parent and last component.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Al Viro8737f3a2008-08-02 22:36:57 -04001793 error = -ENFILE;
1794 filp = get_empty_filp();
1795 if (filp == NULL)
1796 goto exit_parent;
1797 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001798 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001799 nd.intent.open.flags = flag;
1800 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001801 nd.flags &= ~LOOKUP_PARENT;
Al Viro1f36f772009-12-26 10:56:19 -05001802 nd.flags |= LOOKUP_OPEN;
1803 if (open_flag & O_CREAT) {
1804 nd.flags |= LOOKUP_CREATE;
1805 if (open_flag & O_EXCL)
1806 nd.flags |= LOOKUP_EXCL;
1807 }
Al Viro3e297b62010-03-26 12:40:13 -04001808 if (open_flag & O_DIRECTORY)
1809 nd.flags |= LOOKUP_DIRECTORY;
Jan Kara002baee2010-05-13 12:52:57 +02001810 if (!(open_flag & O_NOFOLLOW))
1811 nd.flags |= LOOKUP_FOLLOW;
Al Viro3e297b62010-03-26 12:40:13 -04001812 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
Al Viro806b6812009-12-26 07:16:40 -05001813 while (unlikely(!filp)) { /* trailing symlink */
Al Virodef4af32009-12-26 08:37:05 -05001814 struct path holder;
Al Viro1f36f772009-12-26 10:56:19 -05001815 struct inode *inode = path.dentry->d_inode;
Al Virodef4af32009-12-26 08:37:05 -05001816 void *cookie;
Al Viro806b6812009-12-26 07:16:40 -05001817 error = -ELOOP;
Al Viro1f36f772009-12-26 10:56:19 -05001818 /* S_ISDIR part is a temporary automount kludge */
Jan Kara002baee2010-05-13 12:52:57 +02001819 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
Al Viro1f36f772009-12-26 10:56:19 -05001820 goto exit_dput;
1821 if (count++ == 32)
Al Viro806b6812009-12-26 07:16:40 -05001822 goto exit_dput;
1823 /*
1824 * This is subtle. Instead of calling do_follow_link() we do
1825 * the thing by hands. The reason is that this way we have zero
1826 * link_count and path_walk() (called from ->follow_link)
1827 * honoring LOOKUP_PARENT. After that we have the parent and
1828 * last component, i.e. we are in the same situation as after
1829 * the first path_walk(). Well, almost - if the last component
1830 * is normal we get its copy stored in nd->last.name and we will
1831 * have to putname() it when we are done. Procfs-like symlinks
1832 * just set LAST_BIND.
1833 */
1834 nd.flags |= LOOKUP_PARENT;
1835 error = security_inode_follow_link(path.dentry, &nd);
1836 if (error)
1837 goto exit_dput;
Al Virodef4af32009-12-26 08:37:05 -05001838 error = __do_follow_link(&path, &nd, &cookie);
1839 if (unlikely(error)) {
Al Viro806b6812009-12-26 07:16:40 -05001840 /* nd.path had been dropped */
Al Virodef4af32009-12-26 08:37:05 -05001841 if (!IS_ERR(cookie) && inode->i_op->put_link)
1842 inode->i_op->put_link(path.dentry, &nd, cookie);
1843 path_put(&path);
Al Viro806b6812009-12-26 07:16:40 -05001844 release_open_intent(&nd);
1845 filp = ERR_PTR(error);
1846 goto out;
1847 }
Al Virodef4af32009-12-26 08:37:05 -05001848 holder = path;
Al Viro806b6812009-12-26 07:16:40 -05001849 nd.flags &= ~LOOKUP_PARENT;
Al Viro3e297b62010-03-26 12:40:13 -04001850 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
Al Virodef4af32009-12-26 08:37:05 -05001851 if (inode->i_op->put_link)
1852 inode->i_op->put_link(holder.dentry, &nd, cookie);
1853 path_put(&holder);
Al Viro806b6812009-12-26 07:16:40 -05001854 }
Al Viro10fa8e62009-12-26 07:09:49 -05001855out:
Al Viro2a737872009-04-07 11:49:53 -04001856 if (nd.root.mnt)
1857 path_put(&nd.root);
Al Viro10fa8e62009-12-26 07:09:49 -05001858 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1859 force_reval = 1;
1860 goto reval;
1861 }
1862 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
Al Viro806b6812009-12-26 07:16:40 -05001864exit_dput:
1865 path_put_conditional(&path, &nd);
1866 if (!IS_ERR(nd.intent.open.file))
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001867 release_open_intent(&nd);
Al Viro806b6812009-12-26 07:16:40 -05001868exit_parent:
1869 path_put(&nd.path);
1870 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05001871 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872}
1873
1874/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001875 * filp_open - open file and return file pointer
1876 *
1877 * @filename: path to open
1878 * @flags: open flags as per the open(2) second argument
1879 * @mode: mode for the new file if O_CREAT is set, else ignored
1880 *
1881 * This is the helper to open a file from kernelspace if you really
1882 * have to. But in generally you should not do this, so please move
1883 * along, nothing to see here..
1884 */
1885struct file *filp_open(const char *filename, int flags, int mode)
1886{
Al Viro6e8341a2009-04-06 11:16:22 -04001887 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001888}
1889EXPORT_SYMBOL(filp_open);
1890
1891/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 * lookup_create - lookup a dentry, creating it if it doesn't exist
1893 * @nd: nameidata info
1894 * @is_dir: directory flag
1895 *
1896 * Simple function to lookup and return a dentry and create it
1897 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001898 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001899 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 */
1901struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1902{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001903 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
Jan Blunck4ac91372008-02-14 19:34:32 -08001905 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001906 /*
1907 * Yucky last component or no last component at all?
1908 * (foo/., foo/.., /////)
1909 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 if (nd->last_type != LAST_NORM)
1911 goto fail;
1912 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001913 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001914 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001915
1916 /*
1917 * Do the final lookup.
1918 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001919 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 if (IS_ERR(dentry))
1921 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001922
Al Viroe9baf6e2008-05-15 04:49:12 -04001923 if (dentry->d_inode)
1924 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001925 /*
1926 * Special case - lookup gave negative, but... we had foo/bar/
1927 * From the vfs_mknod() POV we just have a negative dentry -
1928 * all is fine. Let's be bastards - you had / on the end, you've
1929 * been asking for (non-existent) directory. -ENOENT for you.
1930 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001931 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1932 dput(dentry);
1933 dentry = ERR_PTR(-ENOENT);
1934 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001936eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001938 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939fail:
1940 return dentry;
1941}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001942EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1945{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001946 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
1948 if (error)
1949 return error;
1950
1951 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1952 return -EPERM;
1953
Al Viroacfa4382008-12-04 10:06:33 -05001954 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return -EPERM;
1956
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001957 error = devcgroup_inode_mknod(mode, dev);
1958 if (error)
1959 return error;
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 error = security_inode_mknod(dir, dentry, mode, dev);
1962 if (error)
1963 return error;
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001966 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001967 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 return error;
1969}
1970
Dave Hansen463c3192008-02-15 14:37:57 -08001971static int may_mknod(mode_t mode)
1972{
1973 switch (mode & S_IFMT) {
1974 case S_IFREG:
1975 case S_IFCHR:
1976 case S_IFBLK:
1977 case S_IFIFO:
1978 case S_IFSOCK:
1979 case 0: /* zero mode translates to S_IFREG */
1980 return 0;
1981 case S_IFDIR:
1982 return -EPERM;
1983 default:
1984 return -EINVAL;
1985 }
1986}
1987
Heiko Carstens2e4d0922009-01-14 14:14:31 +01001988SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1989 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990{
Al Viro2ad94ae2008-07-21 09:32:51 -04001991 int error;
1992 char *tmp;
1993 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 struct nameidata nd;
1995
1996 if (S_ISDIR(mode))
1997 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Al Viro2ad94ae2008-07-21 09:32:51 -04001999 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002001 return error;
2002
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002004 if (IS_ERR(dentry)) {
2005 error = PTR_ERR(dentry);
2006 goto out_unlock;
2007 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002008 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002009 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002010 error = may_mknod(mode);
2011 if (error)
2012 goto out_dput;
2013 error = mnt_want_write(nd.path.mnt);
2014 if (error)
2015 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002016 error = security_path_mknod(&nd.path, dentry, mode, dev);
2017 if (error)
2018 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002019 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002021 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 break;
2023 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002024 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 new_decode_dev(dev));
2026 break;
2027 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002028 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002031out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002032 mnt_drop_write(nd.path.mnt);
2033out_dput:
2034 dput(dentry);
2035out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002036 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002037 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 putname(tmp);
2039
2040 return error;
2041}
2042
Heiko Carstens3480b252009-01-14 14:14:16 +01002043SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002044{
2045 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2046}
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2049{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002050 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
2052 if (error)
2053 return error;
2054
Al Viroacfa4382008-12-04 10:06:33 -05002055 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 return -EPERM;
2057
2058 mode &= (S_IRWXUGO|S_ISVTX);
2059 error = security_inode_mkdir(dir, dentry, mode);
2060 if (error)
2061 return error;
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002064 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002065 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 return error;
2067}
2068
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002069SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
2071 int error = 0;
2072 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002073 struct dentry *dentry;
2074 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
Al Viro2ad94ae2008-07-21 09:32:51 -04002076 error = user_path_parent(dfd, pathname, &nd, &tmp);
2077 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002078 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Dave Hansen6902d922006-09-30 23:29:01 -07002080 dentry = lookup_create(&nd, 1);
2081 error = PTR_ERR(dentry);
2082 if (IS_ERR(dentry))
2083 goto out_unlock;
2084
Jan Blunck4ac91372008-02-14 19:34:32 -08002085 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002086 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002087 error = mnt_want_write(nd.path.mnt);
2088 if (error)
2089 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002090 error = security_path_mkdir(&nd.path, dentry, mode);
2091 if (error)
2092 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002093 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002094out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002095 mnt_drop_write(nd.path.mnt);
2096out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002097 dput(dentry);
2098out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002099 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002100 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002101 putname(tmp);
2102out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 return error;
2104}
2105
Heiko Carstens3cdad422009-01-14 14:14:22 +01002106SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002107{
2108 return sys_mkdirat(AT_FDCWD, pathname, mode);
2109}
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111/*
2112 * We try to drop the dentry early: we should have
2113 * a usage count of 2 if we're the only user of this
2114 * dentry, and if that is true (possibly after pruning
2115 * the dcache), then we drop the dentry now.
2116 *
2117 * A low-level filesystem can, if it choses, legally
2118 * do a
2119 *
2120 * if (!d_unhashed(dentry))
2121 * return -EBUSY;
2122 *
2123 * if it cannot handle the case of removing a directory
2124 * that is still in use by something else..
2125 */
2126void dentry_unhash(struct dentry *dentry)
2127{
2128 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002129 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 spin_lock(&dcache_lock);
2131 spin_lock(&dentry->d_lock);
2132 if (atomic_read(&dentry->d_count) == 2)
2133 __d_drop(dentry);
2134 spin_unlock(&dentry->d_lock);
2135 spin_unlock(&dcache_lock);
2136}
2137
2138int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2139{
2140 int error = may_delete(dir, dentry, 1);
2141
2142 if (error)
2143 return error;
2144
Al Viroacfa4382008-12-04 10:06:33 -05002145 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 return -EPERM;
2147
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002148 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 dentry_unhash(dentry);
2150 if (d_mountpoint(dentry))
2151 error = -EBUSY;
2152 else {
2153 error = security_inode_rmdir(dir, dentry);
2154 if (!error) {
2155 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002156 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002158 dont_mount(dentry);
2159 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 }
2161 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002162 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 d_delete(dentry);
2165 }
2166 dput(dentry);
2167
2168 return error;
2169}
2170
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002171static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172{
2173 int error = 0;
2174 char * name;
2175 struct dentry *dentry;
2176 struct nameidata nd;
2177
Al Viro2ad94ae2008-07-21 09:32:51 -04002178 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002180 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
2182 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002183 case LAST_DOTDOT:
2184 error = -ENOTEMPTY;
2185 goto exit1;
2186 case LAST_DOT:
2187 error = -EINVAL;
2188 goto exit1;
2189 case LAST_ROOT:
2190 error = -EBUSY;
2191 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002193
2194 nd.flags &= ~LOOKUP_PARENT;
2195
Jan Blunck4ac91372008-02-14 19:34:32 -08002196 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002197 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002199 if (IS_ERR(dentry))
2200 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002201 error = mnt_want_write(nd.path.mnt);
2202 if (error)
2203 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002204 error = security_path_rmdir(&nd.path, dentry);
2205 if (error)
2206 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002207 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002208exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002209 mnt_drop_write(nd.path.mnt);
2210exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002211 dput(dentry);
2212exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002213 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002215 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 putname(name);
2217 return error;
2218}
2219
Heiko Carstens3cdad422009-01-14 14:14:22 +01002220SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002221{
2222 return do_rmdir(AT_FDCWD, pathname);
2223}
2224
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225int vfs_unlink(struct inode *dir, struct dentry *dentry)
2226{
2227 int error = may_delete(dir, dentry, 0);
2228
2229 if (error)
2230 return error;
2231
Al Viroacfa4382008-12-04 10:06:33 -05002232 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 return -EPERM;
2234
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002235 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 if (d_mountpoint(dentry))
2237 error = -EBUSY;
2238 else {
2239 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002240 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002242 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002243 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002246 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2249 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002250 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 }
Robert Love0eeca282005-07-12 17:06:03 -04002253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 return error;
2255}
2256
2257/*
2258 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002259 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 * writeout happening, and we don't want to prevent access to the directory
2261 * while waiting on the I/O.
2262 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002263static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264{
Al Viro2ad94ae2008-07-21 09:32:51 -04002265 int error;
2266 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 struct dentry *dentry;
2268 struct nameidata nd;
2269 struct inode *inode = NULL;
2270
Al Viro2ad94ae2008-07-21 09:32:51 -04002271 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002273 return error;
2274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 error = -EISDIR;
2276 if (nd.last_type != LAST_NORM)
2277 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002278
2279 nd.flags &= ~LOOKUP_PARENT;
2280
Jan Blunck4ac91372008-02-14 19:34:32 -08002281 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002282 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 error = PTR_ERR(dentry);
2284 if (!IS_ERR(dentry)) {
2285 /* Why not before? Because we want correct error value */
2286 if (nd.last.name[nd.last.len])
2287 goto slashes;
2288 inode = dentry->d_inode;
2289 if (inode)
Al Viro7de9c6e2010-10-23 11:11:40 -04002290 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002291 error = mnt_want_write(nd.path.mnt);
2292 if (error)
2293 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002294 error = security_path_unlink(&nd.path, dentry);
2295 if (error)
2296 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002297 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002298exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002299 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 exit2:
2301 dput(dentry);
2302 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002303 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 if (inode)
2305 iput(inode); /* truncate the inode here */
2306exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002307 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 putname(name);
2309 return error;
2310
2311slashes:
2312 error = !dentry->d_inode ? -ENOENT :
2313 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2314 goto exit2;
2315}
2316
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002317SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002318{
2319 if ((flag & ~AT_REMOVEDIR) != 0)
2320 return -EINVAL;
2321
2322 if (flag & AT_REMOVEDIR)
2323 return do_rmdir(dfd, pathname);
2324
2325 return do_unlinkat(dfd, pathname);
2326}
2327
Heiko Carstens3480b252009-01-14 14:14:16 +01002328SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002329{
2330 return do_unlinkat(AT_FDCWD, pathname);
2331}
2332
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002333int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002335 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
2337 if (error)
2338 return error;
2339
Al Viroacfa4382008-12-04 10:06:33 -05002340 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 return -EPERM;
2342
2343 error = security_inode_symlink(dir, dentry, oldname);
2344 if (error)
2345 return error;
2346
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002348 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002349 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 return error;
2351}
2352
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002353SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2354 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355{
Al Viro2ad94ae2008-07-21 09:32:51 -04002356 int error;
2357 char *from;
2358 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002359 struct dentry *dentry;
2360 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361
2362 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002363 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002365
2366 error = user_path_parent(newdfd, newname, &nd, &to);
2367 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002368 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Dave Hansen6902d922006-09-30 23:29:01 -07002370 dentry = lookup_create(&nd, 0);
2371 error = PTR_ERR(dentry);
2372 if (IS_ERR(dentry))
2373 goto out_unlock;
2374
Dave Hansen75c3f292008-02-15 14:37:45 -08002375 error = mnt_want_write(nd.path.mnt);
2376 if (error)
2377 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002378 error = security_path_symlink(&nd.path, dentry, from);
2379 if (error)
2380 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002381 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002382out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002383 mnt_drop_write(nd.path.mnt);
2384out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002385 dput(dentry);
2386out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002387 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002388 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002389 putname(to);
2390out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 putname(from);
2392 return error;
2393}
2394
Heiko Carstens3480b252009-01-14 14:14:16 +01002395SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002396{
2397 return sys_symlinkat(oldname, AT_FDCWD, newname);
2398}
2399
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2401{
2402 struct inode *inode = old_dentry->d_inode;
2403 int error;
2404
2405 if (!inode)
2406 return -ENOENT;
2407
Miklos Szeredia95164d2008-07-30 15:08:48 +02002408 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 if (error)
2410 return error;
2411
2412 if (dir->i_sb != inode->i_sb)
2413 return -EXDEV;
2414
2415 /*
2416 * A link to an append-only or immutable file cannot be created.
2417 */
2418 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2419 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002420 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002422 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 return -EPERM;
2424
2425 error = security_inode_link(old_dentry, dir, new_dentry);
2426 if (error)
2427 return error;
2428
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002429 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002431 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002432 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002433 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 return error;
2435}
2436
2437/*
2438 * Hardlinks are often used in delicate situations. We avoid
2439 * security-related surprises by not following symlinks on the
2440 * newname. --KAB
2441 *
2442 * We don't follow them on the oldname either to be compatible
2443 * with linux 2.0, and to avoid hard-linking to directories
2444 * and other special files. --ADM
2445 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002446SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2447 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
2449 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002450 struct nameidata nd;
2451 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002453 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002455 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002456 return -EINVAL;
2457
Al Viro2d8f3032008-07-22 09:59:21 -04002458 error = user_path_at(olddfd, oldname,
2459 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2460 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002462 return error;
2463
2464 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (error)
2466 goto out;
2467 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002468 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 goto out_release;
2470 new_dentry = lookup_create(&nd, 0);
2471 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002472 if (IS_ERR(new_dentry))
2473 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002474 error = mnt_want_write(nd.path.mnt);
2475 if (error)
2476 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002477 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2478 if (error)
2479 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002480 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002481out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002482 mnt_drop_write(nd.path.mnt);
2483out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002484 dput(new_dentry);
2485out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002486 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002488 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002489 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490out:
Al Viro2d8f3032008-07-22 09:59:21 -04002491 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
2493 return error;
2494}
2495
Heiko Carstens3480b252009-01-14 14:14:16 +01002496SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002497{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002498 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002499}
2500
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501/*
2502 * The worst of all namespace operations - renaming directory. "Perverted"
2503 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2504 * Problems:
2505 * a) we can get into loop creation. Check is done in is_subdir().
2506 * b) race potential - two innocent renames can create a loop together.
2507 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002508 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 * story.
2510 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002511 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 * whether the target exists). Solution: try to be smart with locking
2513 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002514 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 * move will be locked. Thus we can rank directories by the tree
2516 * (ancestors first) and rank all non-directories after them.
2517 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002518 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 * HOWEVER, it relies on the assumption that any object with ->lookup()
2520 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2521 * we'd better make sure that there's no link(2) for them.
2522 * d) some filesystems don't support opened-but-unlinked directories,
2523 * either because of layout or because they are not ready to deal with
2524 * all cases correctly. The latter will be fixed (taking this sort of
2525 * stuff into VFS), but the former is not going away. Solution: the same
2526 * trick as in rmdir().
2527 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002528 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002530 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 * locking].
2532 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002533static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2534 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535{
2536 int error = 0;
2537 struct inode *target;
2538
2539 /*
2540 * If we are going to change the parent - check write permissions,
2541 * we'll need to flip '..'.
2542 */
2543 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002544 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 if (error)
2546 return error;
2547 }
2548
2549 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2550 if (error)
2551 return error;
2552
2553 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04002554 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002555 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2557 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04002558 else {
2559 if (target)
2560 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04002564 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002566 dont_mount(new_dentry);
2567 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002568 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 if (d_unhashed(new_dentry))
2570 d_rehash(new_dentry);
2571 dput(new_dentry);
2572 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002573 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002574 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2575 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 return error;
2577}
2578
Adrian Bunk75c96f82005-05-05 16:16:09 -07002579static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2580 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581{
2582 struct inode *target;
2583 int error;
2584
2585 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2586 if (error)
2587 return error;
2588
2589 dget(new_dentry);
2590 target = new_dentry->d_inode;
2591 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002592 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2594 error = -EBUSY;
2595 else
2596 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2597 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05002598 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04002599 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07002600 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 }
2603 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002604 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 dput(new_dentry);
2606 return error;
2607}
2608
2609int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2610 struct inode *new_dir, struct dentry *new_dentry)
2611{
2612 int error;
2613 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05002614 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
2616 if (old_dentry->d_inode == new_dentry->d_inode)
2617 return 0;
2618
2619 error = may_delete(old_dir, old_dentry, is_dir);
2620 if (error)
2621 return error;
2622
2623 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002624 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 else
2626 error = may_delete(new_dir, new_dentry, is_dir);
2627 if (error)
2628 return error;
2629
Al Viroacfa4382008-12-04 10:06:33 -05002630 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 return -EPERM;
2632
Robert Love0eeca282005-07-12 17:06:03 -04002633 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2634
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 if (is_dir)
2636 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2637 else
2638 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05002639 if (!error)
2640 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002641 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04002642 fsnotify_oldname_free(old_name);
2643
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 return error;
2645}
2646
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002647SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2648 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649{
Al Viro2ad94ae2008-07-21 09:32:51 -04002650 struct dentry *old_dir, *new_dir;
2651 struct dentry *old_dentry, *new_dentry;
2652 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002654 char *from;
2655 char *to;
2656 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
Al Viro2ad94ae2008-07-21 09:32:51 -04002658 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 if (error)
2660 goto exit;
2661
Al Viro2ad94ae2008-07-21 09:32:51 -04002662 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 if (error)
2664 goto exit1;
2665
2666 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002667 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 goto exit2;
2669
Jan Blunck4ac91372008-02-14 19:34:32 -08002670 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 error = -EBUSY;
2672 if (oldnd.last_type != LAST_NORM)
2673 goto exit2;
2674
Jan Blunck4ac91372008-02-14 19:34:32 -08002675 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 if (newnd.last_type != LAST_NORM)
2677 goto exit2;
2678
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002679 oldnd.flags &= ~LOOKUP_PARENT;
2680 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002681 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002682
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 trap = lock_rename(new_dir, old_dir);
2684
Christoph Hellwig49705b72005-11-08 21:35:06 -08002685 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 error = PTR_ERR(old_dentry);
2687 if (IS_ERR(old_dentry))
2688 goto exit3;
2689 /* source must exist */
2690 error = -ENOENT;
2691 if (!old_dentry->d_inode)
2692 goto exit4;
2693 /* unless the source is a directory trailing slashes give -ENOTDIR */
2694 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2695 error = -ENOTDIR;
2696 if (oldnd.last.name[oldnd.last.len])
2697 goto exit4;
2698 if (newnd.last.name[newnd.last.len])
2699 goto exit4;
2700 }
2701 /* source should not be ancestor of target */
2702 error = -EINVAL;
2703 if (old_dentry == trap)
2704 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002705 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 error = PTR_ERR(new_dentry);
2707 if (IS_ERR(new_dentry))
2708 goto exit4;
2709 /* target should not be an ancestor of source */
2710 error = -ENOTEMPTY;
2711 if (new_dentry == trap)
2712 goto exit5;
2713
Dave Hansen9079b1e2008-02-15 14:37:49 -08002714 error = mnt_want_write(oldnd.path.mnt);
2715 if (error)
2716 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002717 error = security_path_rename(&oldnd.path, old_dentry,
2718 &newnd.path, new_dentry);
2719 if (error)
2720 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 error = vfs_rename(old_dir->d_inode, old_dentry,
2722 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002723exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002724 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725exit5:
2726 dput(new_dentry);
2727exit4:
2728 dput(old_dentry);
2729exit3:
2730 unlock_rename(new_dir, old_dir);
2731exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002732 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002733 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002735 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002737exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 return error;
2739}
2740
Heiko Carstensa26eab22009-01-14 14:14:17 +01002741SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002742{
2743 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2744}
2745
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2747{
2748 int len;
2749
2750 len = PTR_ERR(link);
2751 if (IS_ERR(link))
2752 goto out;
2753
2754 len = strlen(link);
2755 if (len > (unsigned) buflen)
2756 len = buflen;
2757 if (copy_to_user(buffer, link, len))
2758 len = -EFAULT;
2759out:
2760 return len;
2761}
2762
2763/*
2764 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2765 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2766 * using) it for any given inode is up to filesystem.
2767 */
2768int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2769{
2770 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002771 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002772 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002773
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002775 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002776 if (IS_ERR(cookie))
2777 return PTR_ERR(cookie);
2778
2779 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2780 if (dentry->d_inode->i_op->put_link)
2781 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2782 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783}
2784
2785int vfs_follow_link(struct nameidata *nd, const char *link)
2786{
2787 return __vfs_follow_link(nd, link);
2788}
2789
2790/* get the link contents into pagecache */
2791static char *page_getlink(struct dentry * dentry, struct page **ppage)
2792{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002793 char *kaddr;
2794 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002796 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002798 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002800 kaddr = kmap(page);
2801 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2802 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803}
2804
2805int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2806{
2807 struct page *page = NULL;
2808 char *s = page_getlink(dentry, &page);
2809 int res = vfs_readlink(dentry,buffer,buflen,s);
2810 if (page) {
2811 kunmap(page);
2812 page_cache_release(page);
2813 }
2814 return res;
2815}
2816
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002817void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002819 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002821 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822}
2823
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002824void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002826 struct page *page = cookie;
2827
2828 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 kunmap(page);
2830 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 }
2832}
2833
Nick Piggin54566b22009-01-04 12:00:53 -08002834/*
2835 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2836 */
2837int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838{
2839 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002840 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002841 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002842 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002844 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2845 if (nofs)
2846 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847
NeilBrown7e53cac2006-03-25 03:07:57 -08002848retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002849 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002850 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002852 goto fail;
2853
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 kaddr = kmap_atomic(page, KM_USER0);
2855 memcpy(kaddr, symname, len-1);
2856 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002857
2858 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2859 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 if (err < 0)
2861 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002862 if (err < len-1)
2863 goto retry;
2864
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 mark_inode_dirty(inode);
2866 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867fail:
2868 return err;
2869}
2870
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002871int page_symlink(struct inode *inode, const char *symname, int len)
2872{
2873 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002874 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002875}
2876
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002877const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 .readlink = generic_readlink,
2879 .follow_link = page_follow_link_light,
2880 .put_link = page_put_link,
2881};
2882
Al Viro2d8f3032008-07-22 09:59:21 -04002883EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884EXPORT_SYMBOL(follow_down);
2885EXPORT_SYMBOL(follow_up);
2886EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2887EXPORT_SYMBOL(getname);
2888EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889EXPORT_SYMBOL(lookup_one_len);
2890EXPORT_SYMBOL(page_follow_link_light);
2891EXPORT_SYMBOL(page_put_link);
2892EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002893EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894EXPORT_SYMBOL(page_symlink);
2895EXPORT_SYMBOL(page_symlink_inode_operations);
2896EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002897EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002898EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002899EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002900EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901EXPORT_SYMBOL(unlock_rename);
2902EXPORT_SYMBOL(vfs_create);
2903EXPORT_SYMBOL(vfs_follow_link);
2904EXPORT_SYMBOL(vfs_link);
2905EXPORT_SYMBOL(vfs_mkdir);
2906EXPORT_SYMBOL(vfs_mknod);
2907EXPORT_SYMBOL(generic_permission);
2908EXPORT_SYMBOL(vfs_readlink);
2909EXPORT_SYMBOL(vfs_rename);
2910EXPORT_SYMBOL(vfs_rmdir);
2911EXPORT_SYMBOL(vfs_symlink);
2912EXPORT_SYMBOL(vfs_unlink);
2913EXPORT_SYMBOL(dentry_unhash);
2914EXPORT_SYMBOL(generic_readlink);