blob: 48e1f60520ea51d64f1802bda2b1d832249e4067 [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
Al Viroe6305c42008-07-15 21:03:57 -0400285 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
288
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800289/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
293 *
294 * Used to check for read/write/execute permissions on an already opened
295 * file.
296 *
297 * Note:
298 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200299 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800300 */
301int file_permission(struct file *file, int mask)
302{
Al Virof419a2e2008-07-22 00:07:17 -0400303 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800304}
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306/*
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
316 *
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
322 */
323
324int get_write_access(struct inode * inode)
325{
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
329 return -ETXTBSY;
330 }
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
333
334 return 0;
335}
336
337int deny_write_access(struct file * file)
338{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800339 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
344 return -ETXTBSY;
345 }
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
348
349 return 0;
350}
351
Jan Blunck1d957f92008-02-14 19:34:35 -0800352/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800353 * path_get - get a reference to a path
354 * @path: path to get the reference to
355 *
356 * Given a path increment the reference count to the dentry and the vfsmount.
357 */
358void path_get(struct path *path)
359{
360 mntget(path->mnt);
361 dget(path->dentry);
362}
363EXPORT_SYMBOL(path_get);
364
365/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800366 * path_put - put a reference to a path
367 * @path: path to put the reference to
368 *
369 * Given a path decrement the reference count to the dentry and the vfsmount.
370 */
371void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Jan Blunck1d957f92008-02-14 19:34:35 -0800373 dput(path->dentry);
374 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Jan Blunck1d957f92008-02-14 19:34:35 -0800376EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Trond Myklebust834f2a42005-10-18 14:20:16 -0700378/**
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
381 */
382void release_open_intent(struct nameidata *nd)
383{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800384 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700385 put_filp(nd->intent.open.file);
386 else
387 fput(nd->intent.open.file);
388}
389
Ian Kentbcdc5e02006-09-27 01:50:44 -0700390static inline struct dentry *
391do_revalidate(struct dentry *dentry, struct nameidata *nd)
392{
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
395 /*
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
400 */
401 if (!status) {
402 if (!d_invalidate(dentry)) {
403 dput(dentry);
404 dentry = NULL;
405 }
406 } else {
407 dput(dentry);
408 dentry = ERR_PTR(status);
409 }
410 }
411 return dentry;
412}
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414/*
Jeff Layton39159de2009-12-07 12:01:50 -0500415 * force_reval_path - force revalidation of a dentry
416 *
417 * In some situations the path walking code will trust dentries without
418 * revalidating them. This causes problems for filesystems that depend on
419 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
420 * (which indicates that it's possible for the dentry to go stale), force
421 * a d_revalidate call before proceeding.
422 *
423 * Returns 0 if the revalidation was successful. If the revalidation fails,
424 * either return the error returned by d_revalidate or -ESTALE if the
425 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
426 * invalidate the dentry. It's up to the caller to handle putting references
427 * to the path if necessary.
428 */
429static int
430force_reval_path(struct path *path, struct nameidata *nd)
431{
432 int status;
433 struct dentry *dentry = path->dentry;
434
435 /*
436 * only check on filesystems where it's possible for the dentry to
437 * become stale. It's assumed that if this flag is set then the
438 * d_revalidate op will also be defined.
439 */
440 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
441 return 0;
442
443 status = dentry->d_op->d_revalidate(dentry, nd);
444 if (status > 0)
445 return 0;
446
447 if (!status) {
448 d_invalidate(dentry);
449 status = -ESTALE;
450 }
451 return status;
452}
453
454/*
Al Virob75b5082009-12-16 01:01:38 -0500455 * Short-cut version of permission(), for calling on directories
456 * during pathname resolution. Combines parts of permission()
457 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 *
459 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500460 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * complete permission check.
462 */
Al Virob75b5082009-12-16 01:01:38 -0500463static int exec_permission(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700465 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700467 if (inode->i_op->permission) {
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700468 ret = inode->i_op->permission(inode, MAY_EXEC);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700469 if (!ret)
470 goto ok;
471 return ret;
472 }
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700473 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
474 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 goto ok;
476
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700477 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 goto ok;
479
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700480 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481ok:
Al Virob77b0642008-07-17 09:37:02 -0400482 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483}
484
Al Viro2a737872009-04-07 11:49:53 -0400485static __always_inline void set_root(struct nameidata *nd)
486{
487 if (!nd->root.mnt) {
488 struct fs_struct *fs = current->fs;
489 read_lock(&fs->lock);
490 nd->root = fs->root;
491 path_get(&nd->root);
492 read_unlock(&fs->lock);
493 }
494}
495
Al Viro6de88d72009-08-09 01:41:57 +0400496static int link_path_walk(const char *, struct nameidata *);
497
Arjan van de Venf1662352006-01-14 13:21:31 -0800498static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (IS_ERR(link))
501 goto fail;
502
503 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400504 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800505 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400506 nd->path = nd->root;
507 path_get(&nd->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100509
Al Virodef4af32009-12-26 08:37:05 -0500510 return link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800512 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return PTR_ERR(link);
514}
515
Jan Blunck1d957f92008-02-14 19:34:35 -0800516static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700517{
518 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800519 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700520 mntput(path->mnt);
521}
522
523static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
524{
Jan Blunck4ac91372008-02-14 19:34:32 -0800525 dput(nd->path.dentry);
Huang Shijie9a229682010-04-02 17:37:13 +0800526 if (nd->path.mnt != path->mnt) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800527 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800528 nd->path.mnt = path->mnt;
529 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800530 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700531}
532
Al Virodef4af32009-12-26 08:37:05 -0500533static __always_inline int
534__do_follow_link(struct path *path, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800535{
536 int error;
Ian Kent051d3812006-03-27 01:14:53 -0800537 struct dentry *dentry = path->dentry;
538
539 touch_atime(path->mnt, dentry);
540 nd_set_link(nd, NULL);
541
Jan Blunck4ac91372008-02-14 19:34:32 -0800542 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800543 path_to_nameidata(path, nd);
544 dget(dentry);
545 }
546 mntget(path->mnt);
Al Viro86acdca12009-12-22 23:45:11 -0500547 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500548 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
549 error = PTR_ERR(*p);
550 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800551 char *s = nd_get_link(nd);
552 error = 0;
553 if (s)
554 error = __vfs_follow_link(nd, s);
Jeff Layton39159de2009-12-07 12:01:50 -0500555 else if (nd->last_type == LAST_BIND) {
556 error = force_reval_path(&nd->path, nd);
557 if (error)
558 path_put(&nd->path);
559 }
Ian Kent051d3812006-03-27 01:14:53 -0800560 }
Ian Kent051d3812006-03-27 01:14:53 -0800561 return error;
562}
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564/*
565 * This limits recursive symlink follows to 8, while
566 * limiting consecutive symlinks to 40.
567 *
568 * Without that kind of total limit, nasty chains of consecutive
569 * symlinks can cause almost arbitrarily long lookups.
570 */
Al Viro90ebe562005-06-06 13:35:58 -0700571static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Al Virodef4af32009-12-26 08:37:05 -0500573 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 int err = -ELOOP;
575 if (current->link_count >= MAX_NESTED_LINKS)
576 goto loop;
577 if (current->total_link_count >= 40)
578 goto loop;
579 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
580 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700581 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 if (err)
583 goto loop;
584 current->link_count++;
585 current->total_link_count++;
586 nd->depth++;
Al Virodef4af32009-12-26 08:37:05 -0500587 err = __do_follow_link(path, nd, &cookie);
588 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
589 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
Al Viro258fa992009-08-09 01:32:02 +0400590 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 current->link_count--;
592 nd->depth--;
593 return err;
594loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800595 path_put_conditional(path, nd);
596 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return err;
598}
599
Al Virobab77eb2009-04-18 03:26:48 -0400600int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
602 struct vfsmount *parent;
603 struct dentry *mountpoint;
604 spin_lock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400605 parent = path->mnt->mnt_parent;
606 if (parent == path->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 spin_unlock(&vfsmount_lock);
608 return 0;
609 }
610 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400611 mountpoint = dget(path->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 spin_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400613 dput(path->dentry);
614 path->dentry = mountpoint;
615 mntput(path->mnt);
616 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 return 1;
618}
619
620/* no need for dcache_lock, as serialization is taken care in
621 * namespace.c
622 */
Al Viro463ffb22005-06-06 13:36:05 -0700623static int __follow_mount(struct path *path)
624{
625 int res = 0;
626 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400627 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700628 if (!mounted)
629 break;
630 dput(path->dentry);
631 if (res)
632 mntput(path->mnt);
633 path->mnt = mounted;
634 path->dentry = dget(mounted->mnt_root);
635 res = 1;
636 }
637 return res;
638}
639
Al Viro79ed0222009-04-18 13:59:41 -0400640static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Al Viro79ed0222009-04-18 13:59:41 -0400642 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400643 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 if (!mounted)
645 break;
Al Viro79ed0222009-04-18 13:59:41 -0400646 dput(path->dentry);
647 mntput(path->mnt);
648 path->mnt = mounted;
649 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651}
652
653/* no need for dcache_lock, as serialization is taken care in
654 * namespace.c
655 */
Al Viro9393bd02009-04-18 13:58:15 -0400656int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
658 struct vfsmount *mounted;
659
Al Viro1c755af2009-04-18 14:06:57 -0400660 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400662 dput(path->dentry);
663 mntput(path->mnt);
664 path->mnt = mounted;
665 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return 1;
667 }
668 return 0;
669}
670
Arjan van de Venf1662352006-01-14 13:21:31 -0800671static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
Al Viro2a737872009-04-07 11:49:53 -0400673 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800676 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Al Viro2a737872009-04-07 11:49:53 -0400678 if (nd->path.dentry == nd->root.dentry &&
679 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 break;
681 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800682 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -0500683 /* rare case of legitimate dget_parent()... */
684 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 dput(old);
686 break;
687 }
Al Viro3088dd72010-01-30 15:47:29 -0500688 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 }
Al Viro79ed0222009-04-18 13:59:41 -0400691 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694/*
695 * It's more convoluted than I'd like it to be, but... it's still fairly
696 * small and for now I'd prefer to have fast path as straight as possible.
697 * It _is_ time-critical.
698 */
699static int do_lookup(struct nameidata *nd, struct qstr *name,
700 struct path *path)
701{
Jan Blunck4ac91372008-02-14 19:34:32 -0800702 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400703 struct dentry *dentry, *parent;
704 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400705 /*
706 * See if the low-level filesystem might want
707 * to use its own hash..
708 */
709 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
710 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
711 if (err < 0)
712 return err;
713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Al Viro3cac2602009-08-13 18:27:43 +0400715 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (!dentry)
717 goto need_lookup;
718 if (dentry->d_op && dentry->d_op->d_revalidate)
719 goto need_revalidate;
720done:
721 path->mnt = mnt;
722 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700723 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return 0;
725
726need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400727 parent = nd->path.dentry;
728 dir = parent->d_inode;
729
730 mutex_lock(&dir->i_mutex);
731 /*
732 * First re-do the cached lookup just in case it was created
733 * while we waited for the directory semaphore..
734 *
735 * FIXME! This could use version numbering or similar to
736 * avoid unnecessary cache lookups.
737 *
738 * The "dcache_lock" is purely to protect the RCU list walker
739 * from concurrent renames at this point (we mustn't get false
740 * negatives from the RCU list walk here, unlike the optimistic
741 * fast walk).
742 *
743 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
744 */
745 dentry = d_lookup(parent, name);
746 if (!dentry) {
747 struct dentry *new;
748
749 /* Don't create child dentry for a dead directory. */
750 dentry = ERR_PTR(-ENOENT);
751 if (IS_DEADDIR(dir))
752 goto out_unlock;
753
754 new = d_alloc(parent, name);
755 dentry = ERR_PTR(-ENOMEM);
756 if (new) {
757 dentry = dir->i_op->lookup(dir, new, nd);
758 if (dentry)
759 dput(new);
760 else
761 dentry = new;
762 }
763out_unlock:
764 mutex_unlock(&dir->i_mutex);
765 if (IS_ERR(dentry))
766 goto fail;
767 goto done;
768 }
769
770 /*
771 * Uhhuh! Nasty case: the cache was re-populated while
772 * we waited on the semaphore. Need to revalidate.
773 */
774 mutex_unlock(&dir->i_mutex);
775 if (dentry->d_op && dentry->d_op->d_revalidate) {
776 dentry = do_revalidate(dentry, nd);
777 if (!dentry)
778 dentry = ERR_PTR(-ENOENT);
779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 if (IS_ERR(dentry))
781 goto fail;
782 goto done;
783
784need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700785 dentry = do_revalidate(dentry, nd);
786 if (!dentry)
787 goto need_lookup;
788 if (IS_ERR(dentry))
789 goto fail;
790 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792fail:
793 return PTR_ERR(dentry);
794}
795
796/*
Al Viroac278a92010-02-16 18:09:36 +0000797 * This is a temporary kludge to deal with "automount" symlinks; proper
798 * solution is to trigger them on follow_mount(), so that do_lookup()
799 * would DTRT. To be killed before 2.6.34-final.
800 */
801static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
802{
803 return inode && unlikely(inode->i_op->follow_link) &&
804 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
805}
806
807/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100809 * This is the basic name resolution function, turning a pathname into
810 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100812 * Returns 0 and nd will have valid dentry and mnt on success.
813 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 */
Al Viro6de88d72009-08-09 01:41:57 +0400815static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
817 struct path next;
818 struct inode *inode;
819 int err;
820 unsigned int lookup_flags = nd->flags;
821
822 while (*name=='/')
823 name++;
824 if (!*name)
825 goto return_reval;
826
Jan Blunck4ac91372008-02-14 19:34:32 -0800827 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800829 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 /* At this point we know we have a real path component. */
832 for(;;) {
833 unsigned long hash;
834 struct qstr this;
835 unsigned int c;
836
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700837 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500838 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (err)
840 break;
841
842 this.name = name;
843 c = *(const unsigned char *)name;
844
845 hash = init_name_hash();
846 do {
847 name++;
848 hash = partial_name_hash(c, hash);
849 c = *(const unsigned char *)name;
850 } while (c && (c != '/'));
851 this.len = name - (const char *) this.name;
852 this.hash = end_name_hash(hash);
853
854 /* remove trailing slashes? */
855 if (!c)
856 goto last_component;
857 while (*++name == '/');
858 if (!*name)
859 goto last_with_slashes;
860
861 /*
862 * "." and ".." are special - ".." especially so because it has
863 * to be able to know about the current root directory and
864 * parent relationships.
865 */
866 if (this.name[0] == '.') switch (this.len) {
867 default:
868 break;
869 case 2:
870 if (this.name[1] != '.')
871 break;
Al Viro58c465e2005-06-06 13:36:13 -0700872 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800873 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 /* fallthrough */
875 case 1:
876 continue;
877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 /* This does the actual lookups.. */
879 err = do_lookup(nd, &this, &next);
880 if (err)
881 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
883 err = -ENOENT;
884 inode = next.dentry->d_inode;
885 if (!inode)
886 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700889 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (err)
891 goto return_err;
892 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800893 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 if (!inode)
895 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700896 } else
897 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 err = -ENOTDIR;
899 if (!inode->i_op->lookup)
900 break;
901 continue;
902 /* here ends the main loop */
903
904last_with_slashes:
905 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
906last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800907 /* Clear LOOKUP_CONTINUE iff it was previously unset */
908 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (lookup_flags & LOOKUP_PARENT)
910 goto lookup_parent;
911 if (this.name[0] == '.') switch (this.len) {
912 default:
913 break;
914 case 2:
915 if (this.name[1] != '.')
916 break;
Al Viro58c465e2005-06-06 13:36:13 -0700917 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800918 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 /* fallthrough */
920 case 1:
921 goto return_reval;
922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 err = do_lookup(nd, &this, &next);
924 if (err)
925 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 inode = next.dentry->d_inode;
Al Viroac278a92010-02-16 18:09:36 +0000927 if (follow_on_final(inode, lookup_flags)) {
Al Viro90ebe562005-06-06 13:35:58 -0700928 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 if (err)
930 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800931 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700932 } else
933 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 err = -ENOENT;
935 if (!inode)
936 break;
937 if (lookup_flags & LOOKUP_DIRECTORY) {
938 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500939 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 break;
941 }
942 goto return_base;
943lookup_parent:
944 nd->last = this;
945 nd->last_type = LAST_NORM;
946 if (this.name[0] != '.')
947 goto return_base;
948 if (this.len == 1)
949 nd->last_type = LAST_DOT;
950 else if (this.len == 2 && this.name[1] == '.')
951 nd->last_type = LAST_DOTDOT;
952 else
953 goto return_base;
954return_reval:
955 /*
956 * We bypassed the ordinary revalidation routines.
957 * We may need to check the cached dentry for staleness.
958 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800959 if (nd->path.dentry && nd->path.dentry->d_sb &&
960 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 err = -ESTALE;
962 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800963 if (!nd->path.dentry->d_op->d_revalidate(
964 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 break;
966 }
967return_base:
968 return 0;
969out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800970 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 break;
972 }
Jan Blunck1d957f92008-02-14 19:34:35 -0800973 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974return_err:
975 return err;
976}
977
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800978static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Al Viro6de88d72009-08-09 01:41:57 +0400980 struct path save = nd->path;
981 int result;
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +0400984
985 /* make sure the stuff we saved doesn't go away */
986 path_get(&save);
987
988 result = link_path_walk(name, nd);
989 if (result == -ESTALE) {
990 /* nd->path had been dropped */
991 current->total_link_count = 0;
992 nd->path = save;
993 path_get(&nd->path);
994 nd->flags |= LOOKUP_REVAL;
995 result = link_path_walk(name, nd);
996 }
997
998 path_put(&save);
999
1000 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}
1002
Al Viro9b4a9b12009-04-07 11:44:16 -04001003static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001005 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001006 int fput_needed;
1007 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1010 nd->flags = flags;
1011 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001012 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001015 set_root(nd);
1016 nd->path = nd->root;
1017 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001018 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001019 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001020 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001021 nd->path = fs->pwd;
1022 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001023 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001024 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001025 struct dentry *dentry;
1026
1027 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001028 retval = -EBADF;
1029 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001030 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001031
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001032 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001033
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001034 retval = -ENOTDIR;
1035 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001036 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001037
1038 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001039 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001040 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001041
Jan Blunck5dd784d2008-02-14 19:34:38 -08001042 nd->path = file->f_path;
1043 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001044
1045 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001047 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001048
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001049fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001050 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001051out_fail:
1052 return retval;
1053}
1054
1055/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1056static int do_path_lookup(int dfd, const char *name,
1057 unsigned int flags, struct nameidata *nd)
1058{
1059 int retval = path_init(dfd, name, flags, nd);
1060 if (!retval)
1061 retval = path_walk(name, nd);
1062 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1063 nd->path.dentry->d_inode))
1064 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001065 if (nd->root.mnt) {
1066 path_put(&nd->root);
1067 nd->root.mnt = NULL;
1068 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001069 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070}
1071
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001072int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001073 struct nameidata *nd)
1074{
1075 return do_path_lookup(AT_FDCWD, name, flags, nd);
1076}
1077
Al Virod1811462008-08-02 00:49:18 -04001078int kern_path(const char *name, unsigned int flags, struct path *path)
1079{
1080 struct nameidata nd;
1081 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1082 if (!res)
1083 *path = nd.path;
1084 return res;
1085}
1086
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001087/**
1088 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1089 * @dentry: pointer to dentry of the base directory
1090 * @mnt: pointer to vfs mount of the base directory
1091 * @name: pointer to file name
1092 * @flags: lookup flags
1093 * @nd: pointer to nameidata
1094 */
1095int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1096 const char *name, unsigned int flags,
1097 struct nameidata *nd)
1098{
1099 int retval;
1100
1101 /* same as do_path_lookup */
1102 nd->last_type = LAST_ROOT;
1103 nd->flags = flags;
1104 nd->depth = 0;
1105
Jan Blunckc8e7f442008-06-09 16:40:35 -07001106 nd->path.dentry = dentry;
1107 nd->path.mnt = mnt;
1108 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001109 nd->root = nd->path;
1110 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001111
1112 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001113 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1114 nd->path.dentry->d_inode))
1115 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001116
Al Viro5b857112009-04-07 11:53:49 -04001117 path_put(&nd->root);
1118 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001119
Al Viro2a737872009-04-07 11:49:53 -04001120 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001121}
1122
Christoph Hellwigeead1912007-10-16 23:25:38 -07001123static struct dentry *__lookup_hash(struct qstr *name,
1124 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
James Morris057f6c02007-04-26 00:12:05 -07001126 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct inode *inode;
1128 int err;
1129
1130 inode = base->d_inode;
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
Al Viro6e6b1bd2009-08-13 23:38:37 +04001143 dentry = __d_lookup(base, name);
1144
1145 /* lockess __d_lookup may fail due to concurrent d_move()
1146 * in some unrelated directory, so try with d_lookup
1147 */
1148 if (!dentry)
1149 dentry = d_lookup(base, name);
1150
1151 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1152 dentry = do_revalidate(dentry, nd);
1153
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001155 struct dentry *new;
1156
1157 /* Don't create child dentry for a dead directory. */
1158 dentry = ERR_PTR(-ENOENT);
1159 if (IS_DEADDIR(inode))
1160 goto out;
1161
1162 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 dentry = ERR_PTR(-ENOMEM);
1164 if (!new)
1165 goto out;
1166 dentry = inode->i_op->lookup(inode, new, nd);
1167 if (!dentry)
1168 dentry = new;
1169 else
1170 dput(new);
1171 }
1172out:
1173 return dentry;
1174}
1175
James Morris057f6c02007-04-26 00:12:05 -07001176/*
1177 * Restricted form of lookup. Doesn't follow links, single-component only,
1178 * needs parent already locked. Doesn't follow mounts.
1179 * SMP-safe.
1180 */
Adrian Bunka244e162006-03-31 02:32:11 -08001181static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001183 int err;
1184
Al Virob75b5082009-12-16 01:01:38 -05001185 err = exec_permission(nd->path.dentry->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001186 if (err)
1187 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001188 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190
Christoph Hellwigeead1912007-10-16 23:25:38 -07001191static int __lookup_one_len(const char *name, struct qstr *this,
1192 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193{
1194 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 unsigned int c;
1196
James Morris057f6c02007-04-26 00:12:05 -07001197 this->name = name;
1198 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001200 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 hash = init_name_hash();
1203 while (len--) {
1204 c = *(const unsigned char *)name++;
1205 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001206 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 hash = partial_name_hash(c, hash);
1208 }
James Morris057f6c02007-04-26 00:12:05 -07001209 this->hash = end_name_hash(hash);
1210 return 0;
1211}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Christoph Hellwigeead1912007-10-16 23:25:38 -07001213/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001214 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001215 * @name: pathname component to lookup
1216 * @base: base directory to lookup from
1217 * @len: maximum length @len should be interpreted to
1218 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001219 * Note that this routine is purely a helper for filesystem usage and should
1220 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001221 * nameidata argument is passed to the filesystem methods and a filesystem
1222 * using this helper needs to be prepared for that.
1223 */
James Morris057f6c02007-04-26 00:12:05 -07001224struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1225{
1226 int err;
1227 struct qstr this;
1228
David Woodhouse2f9092e2009-04-20 23:18:37 +01001229 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1230
James Morris057f6c02007-04-26 00:12:05 -07001231 err = __lookup_one_len(name, &this, base, len);
1232 if (err)
1233 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001234
Al Virob75b5082009-12-16 01:01:38 -05001235 err = exec_permission(base->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001236 if (err)
1237 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001238 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001239}
1240
Al Viro2d8f3032008-07-22 09:59:21 -04001241int user_path_at(int dfd, const char __user *name, unsigned flags,
1242 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Al Viro2d8f3032008-07-22 09:59:21 -04001244 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 char *tmp = getname(name);
1246 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001248
1249 BUG_ON(flags & LOOKUP_PARENT);
1250
1251 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001253 if (!err)
1254 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256 return err;
1257}
1258
Al Viro2ad94ae2008-07-21 09:32:51 -04001259static int user_path_parent(int dfd, const char __user *path,
1260 struct nameidata *nd, char **name)
1261{
1262 char *s = getname(path);
1263 int error;
1264
1265 if (IS_ERR(s))
1266 return PTR_ERR(s);
1267
1268 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1269 if (error)
1270 putname(s);
1271 else
1272 *name = s;
1273
1274 return error;
1275}
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277/*
1278 * It's inline, so penalty for filesystems that don't use sticky bit is
1279 * minimal.
1280 */
1281static inline int check_sticky(struct inode *dir, struct inode *inode)
1282{
David Howellsda9592e2008-11-14 10:39:05 +11001283 uid_t fsuid = current_fsuid();
1284
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 if (!(dir->i_mode & S_ISVTX))
1286 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001287 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001289 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 return 0;
1291 return !capable(CAP_FOWNER);
1292}
1293
1294/*
1295 * Check whether we can remove a link victim from directory dir, check
1296 * whether the type of victim is right.
1297 * 1. We can't do it if dir is read-only (done in permission())
1298 * 2. We should have write and exec permissions on dir
1299 * 3. We can't remove anything from append-only dir
1300 * 4. We can't do anything with immutable dir (done in permission())
1301 * 5. If the sticky bit on dir is set we should either
1302 * a. be owner of dir, or
1303 * b. be owner of victim, or
1304 * c. have CAP_FOWNER capability
1305 * 6. If the victim is append-only or immutable we can't do antyhing with
1306 * links pointing to it.
1307 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1308 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1309 * 9. We can't remove a root or mountpoint.
1310 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1311 * nfs_async_unlink().
1312 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001313static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314{
1315 int error;
1316
1317 if (!victim->d_inode)
1318 return -ENOENT;
1319
1320 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001321 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Al Virof419a2e2008-07-22 00:07:17 -04001323 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 if (error)
1325 return error;
1326 if (IS_APPEND(dir))
1327 return -EPERM;
1328 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001329 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 return -EPERM;
1331 if (isdir) {
1332 if (!S_ISDIR(victim->d_inode->i_mode))
1333 return -ENOTDIR;
1334 if (IS_ROOT(victim))
1335 return -EBUSY;
1336 } else if (S_ISDIR(victim->d_inode->i_mode))
1337 return -EISDIR;
1338 if (IS_DEADDIR(dir))
1339 return -ENOENT;
1340 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1341 return -EBUSY;
1342 return 0;
1343}
1344
1345/* Check whether we can create an object with dentry child in directory
1346 * dir.
1347 * 1. We can't do it if child already exists (open has special treatment for
1348 * this case, but since we are inlined it's OK)
1349 * 2. We can't do it if dir is read-only (done in permission())
1350 * 3. We should have write and exec permissions on dir
1351 * 4. We can't do it if dir is immutable (done in permission())
1352 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001353static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354{
1355 if (child->d_inode)
1356 return -EEXIST;
1357 if (IS_DEADDIR(dir))
1358 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001359 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360}
1361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362/*
1363 * p1 and p2 should be directories on the same fs.
1364 */
1365struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1366{
1367 struct dentry *p;
1368
1369 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001370 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 return NULL;
1372 }
1373
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001374 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001376 p = d_ancestor(p2, p1);
1377 if (p) {
1378 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1379 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1380 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
1382
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001383 p = d_ancestor(p1, p2);
1384 if (p) {
1385 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1386 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1387 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 }
1389
Ingo Molnarf2eace22006-07-03 00:25:05 -07001390 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1391 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 return NULL;
1393}
1394
1395void unlock_rename(struct dentry *p1, struct dentry *p2)
1396{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001397 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001399 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001400 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
1402}
1403
1404int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1405 struct nameidata *nd)
1406{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001407 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 if (error)
1410 return error;
1411
Al Viroacfa4382008-12-04 10:06:33 -05001412 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return -EACCES; /* shouldn't it be ENOSYS? */
1414 mode &= S_IALLUGO;
1415 mode |= S_IFREG;
1416 error = security_inode_create(dir, dentry, mode);
1417 if (error)
1418 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001420 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001421 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 return error;
1423}
1424
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001425int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001427 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 struct inode *inode = dentry->d_inode;
1429 int error;
1430
1431 if (!inode)
1432 return -ENOENT;
1433
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001434 switch (inode->i_mode & S_IFMT) {
1435 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001437 case S_IFDIR:
1438 if (acc_mode & MAY_WRITE)
1439 return -EISDIR;
1440 break;
1441 case S_IFBLK:
1442 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001443 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001445 /*FALLTHRU*/
1446 case S_IFIFO:
1447 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001449 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001450 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001451
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001452 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001453 if (error)
1454 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 /*
1457 * An append-only file must be opened in append mode for writing.
1458 */
1459 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001460 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001461 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001463 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 }
1465
1466 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001467 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1468 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470 /*
1471 * Ensure there are no outstanding leases on the file.
1472 */
Al Virob65a9cf2009-12-16 06:27:40 -05001473 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001474}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Al Viro7715b522009-12-16 03:54:00 -05001476static int handle_truncate(struct path *path)
1477{
1478 struct inode *inode = path->dentry->d_inode;
1479 int error = get_write_access(inode);
1480 if (error)
1481 return error;
1482 /*
1483 * Refuse to truncate files with mandatory locks held on them.
1484 */
1485 error = locks_verify_locked(inode);
1486 if (!error)
1487 error = security_path_truncate(path, 0,
1488 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1489 if (!error) {
1490 error = do_truncate(path->dentry, 0,
1491 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1492 NULL);
1493 }
1494 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001495 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496}
1497
Dave Hansend57999e2008-02-15 14:37:27 -08001498/*
1499 * Be careful about ever adding any more callers of this
1500 * function. Its flags must be in the namei format, not
1501 * what get passed to sys_open().
1502 */
1503static int __open_namei_create(struct nameidata *nd, struct path *path,
Al Viro8737c932009-12-24 06:47:55 -05001504 int open_flag, int mode)
Dave Hansenaab520e2006-09-30 23:29:02 -07001505{
1506 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001507 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001508
1509 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001510 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001511 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1512 if (error)
1513 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001514 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001515out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001516 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001517 dput(nd->path.dentry);
1518 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001519 if (error)
1520 return error;
1521 /* Don't check for write permission, don't truncate */
Al Viro8737c932009-12-24 06:47:55 -05001522 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001523}
1524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525/*
Dave Hansend57999e2008-02-15 14:37:27 -08001526 * Note that while the flag value (low two bits) for sys_open means:
1527 * 00 - read-only
1528 * 01 - write-only
1529 * 10 - read-write
1530 * 11 - special
1531 * it is changed into
1532 * 00 - no permissions needed
1533 * 01 - read-permission
1534 * 10 - write-permission
1535 * 11 - read-write
1536 * for the internal routines (ie open_namei()/follow_link() etc)
1537 * This is more logical, and also allows the 00 "no perm needed"
1538 * to be used for symlinks (where the permissions are checked
1539 * later).
1540 *
1541*/
1542static inline int open_to_namei_flags(int flag)
1543{
1544 if ((flag+1) & O_ACCMODE)
1545 flag++;
1546 return flag;
1547}
1548
Al Viro7715b522009-12-16 03:54:00 -05001549static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001550{
1551 /*
1552 * We'll never write to the fs underlying
1553 * a device file.
1554 */
1555 if (special_file(inode->i_mode))
1556 return 0;
1557 return (flag & O_TRUNC);
1558}
1559
Al Viro648fa862009-12-24 01:26:48 -05001560static struct file *finish_open(struct nameidata *nd,
Al Viro9a661792009-12-24 06:49:47 -05001561 int open_flag, int acc_mode)
Al Viro648fa862009-12-24 01:26:48 -05001562{
1563 struct file *filp;
1564 int will_truncate;
1565 int error;
1566
Al Viro9a661792009-12-24 06:49:47 -05001567 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
Al Viro648fa862009-12-24 01:26:48 -05001568 if (will_truncate) {
1569 error = mnt_want_write(nd->path.mnt);
1570 if (error)
1571 goto exit;
1572 }
1573 error = may_open(&nd->path, acc_mode, open_flag);
1574 if (error) {
1575 if (will_truncate)
1576 mnt_drop_write(nd->path.mnt);
1577 goto exit;
1578 }
1579 filp = nameidata_to_filp(nd);
1580 if (!IS_ERR(filp)) {
1581 error = ima_file_check(filp, acc_mode);
1582 if (error) {
1583 fput(filp);
1584 filp = ERR_PTR(error);
1585 }
1586 }
1587 if (!IS_ERR(filp)) {
Al Viro648fa862009-12-24 01:26:48 -05001588 if (will_truncate) {
1589 error = handle_truncate(&nd->path);
1590 if (error) {
1591 fput(filp);
1592 filp = ERR_PTR(error);
1593 }
1594 }
1595 }
1596 /*
1597 * It is now safe to drop the mnt write
1598 * because the filp has had a write taken
1599 * on its behalf.
1600 */
1601 if (will_truncate)
1602 mnt_drop_write(nd->path.mnt);
1603 return filp;
1604
1605exit:
1606 if (!IS_ERR(nd->intent.open.file))
1607 release_open_intent(nd);
1608 path_put(&nd->path);
1609 return ERR_PTR(error);
1610}
1611
Al Virofb1cc552009-12-24 01:58:28 -05001612static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viro5b369df2009-12-24 06:51:13 -05001613 int open_flag, int acc_mode,
Al Viro3e297b62010-03-26 12:40:13 -04001614 int mode, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05001615{
Al Viroa1e28032009-12-24 02:12:06 -05001616 struct dentry *dir = nd->path.dentry;
Al Virofb1cc552009-12-24 01:58:28 -05001617 struct file *filp;
Al Viro1f36f772009-12-26 10:56:19 -05001618 int error = -EISDIR;
Al Virofb1cc552009-12-24 01:58:28 -05001619
Al Viro1f36f772009-12-26 10:56:19 -05001620 switch (nd->last_type) {
1621 case LAST_DOTDOT:
1622 follow_dotdot(nd);
1623 dir = nd->path.dentry;
1624 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1625 if (!dir->d_op->d_revalidate(dir, nd)) {
1626 error = -ESTALE;
1627 goto exit;
1628 }
1629 }
1630 /* fallthrough */
1631 case LAST_DOT:
1632 case LAST_ROOT:
1633 if (open_flag & O_CREAT)
1634 goto exit;
1635 /* fallthrough */
1636 case LAST_BIND:
1637 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05001638 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05001639 }
Al Viro67ee3ad2009-12-26 07:01:01 -05001640
Al Viro1f36f772009-12-26 10:56:19 -05001641 /* trailing slashes? */
1642 if (nd->last.name[nd->last.len]) {
1643 if (open_flag & O_CREAT)
1644 goto exit;
Jan Kara002baee2010-05-13 12:52:57 +02001645 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
Al Viro1f36f772009-12-26 10:56:19 -05001646 }
Al Viroa2c36b42009-12-24 03:39:50 -05001647
Al Viro1f36f772009-12-26 10:56:19 -05001648 /* just plain open? */
1649 if (!(open_flag & O_CREAT)) {
1650 error = do_lookup(nd, &nd->last, path);
1651 if (error)
1652 goto exit;
1653 error = -ENOENT;
1654 if (!path->dentry->d_inode)
1655 goto exit_dput;
1656 if (path->dentry->d_inode->i_op->follow_link)
1657 return NULL;
1658 error = -ENOTDIR;
Al Viro3e297b62010-03-26 12:40:13 -04001659 if (nd->flags & LOOKUP_DIRECTORY) {
1660 if (!path->dentry->d_inode->i_op->lookup)
1661 goto exit_dput;
1662 }
Al Viro1f36f772009-12-26 10:56:19 -05001663 path_to_nameidata(path, nd);
1664 audit_inode(pathname, nd->path.dentry);
1665 goto ok;
1666 }
1667
1668 /* OK, it's O_CREAT */
Al Viroa1e28032009-12-24 02:12:06 -05001669 mutex_lock(&dir->d_inode->i_mutex);
1670
1671 path->dentry = lookup_hash(nd);
1672 path->mnt = nd->path.mnt;
1673
Al Virofb1cc552009-12-24 01:58:28 -05001674 error = PTR_ERR(path->dentry);
1675 if (IS_ERR(path->dentry)) {
1676 mutex_unlock(&dir->d_inode->i_mutex);
1677 goto exit;
1678 }
1679
1680 if (IS_ERR(nd->intent.open.file)) {
1681 error = PTR_ERR(nd->intent.open.file);
1682 goto exit_mutex_unlock;
1683 }
1684
1685 /* Negative dentry, just create the file */
1686 if (!path->dentry->d_inode) {
1687 /*
1688 * This write is needed to ensure that a
1689 * ro->rw transition does not occur between
1690 * the time when the file is created and when
1691 * a permanent write count is taken through
1692 * the 'struct file' in nameidata_to_filp().
1693 */
1694 error = mnt_want_write(nd->path.mnt);
1695 if (error)
1696 goto exit_mutex_unlock;
1697 error = __open_namei_create(nd, path, open_flag, mode);
1698 if (error) {
1699 mnt_drop_write(nd->path.mnt);
1700 goto exit;
1701 }
1702 filp = nameidata_to_filp(nd);
1703 mnt_drop_write(nd->path.mnt);
Al Virofb1cc552009-12-24 01:58:28 -05001704 if (!IS_ERR(filp)) {
1705 error = ima_file_check(filp, acc_mode);
1706 if (error) {
1707 fput(filp);
1708 filp = ERR_PTR(error);
1709 }
1710 }
1711 return filp;
1712 }
1713
1714 /*
1715 * It already exists.
1716 */
1717 mutex_unlock(&dir->d_inode->i_mutex);
1718 audit_inode(pathname, path->dentry);
1719
1720 error = -EEXIST;
Al Viro5b369df2009-12-24 06:51:13 -05001721 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05001722 goto exit_dput;
1723
1724 if (__follow_mount(path)) {
1725 error = -ELOOP;
Al Viro5b369df2009-12-24 06:51:13 -05001726 if (open_flag & O_NOFOLLOW)
Al Virofb1cc552009-12-24 01:58:28 -05001727 goto exit_dput;
1728 }
1729
1730 error = -ENOENT;
1731 if (!path->dentry->d_inode)
1732 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05001733
1734 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05001735 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05001736
1737 path_to_nameidata(path, nd);
1738 error = -EISDIR;
1739 if (S_ISDIR(path->dentry->d_inode->i_mode))
1740 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05001741ok:
Al Viro9a661792009-12-24 06:49:47 -05001742 filp = finish_open(nd, open_flag, acc_mode);
Al Virofb1cc552009-12-24 01:58:28 -05001743 return filp;
1744
1745exit_mutex_unlock:
1746 mutex_unlock(&dir->d_inode->i_mutex);
1747exit_dput:
1748 path_put_conditional(path, nd);
1749exit:
1750 if (!IS_ERR(nd->intent.open.file))
1751 release_open_intent(nd);
Al Virofb1cc552009-12-24 01:58:28 -05001752 path_put(&nd->path);
1753 return ERR_PTR(error);
1754}
1755
Dave Hansend57999e2008-02-15 14:37:27 -08001756/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001757 * Note that the low bits of the passed in "open_flag"
1758 * are not the same as in the local variable "flag". See
1759 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001761struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001762 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001764 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001765 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001766 int error;
Al Viro9850c052010-01-13 15:01:15 -05001767 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 int count = 0;
Dave Hansend57999e2008-02-15 14:37:27 -08001769 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001770 int force_reval = 0;
Al Viro1f36f772009-12-26 10:56:19 -05001771
1772 if (!(open_flag & O_CREAT))
1773 mode = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001775 /*
1776 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1777 * check for O_DSYNC if the need any syncing at all we enforce it's
1778 * always set instead of having to deal with possibly weird behaviour
1779 * for malicious applications setting only __O_SYNC.
1780 */
1781 if (open_flag & __O_SYNC)
1782 open_flag |= O_DSYNC;
1783
Al Viro6e8341a2009-04-06 11:16:22 -04001784 if (!acc_mode)
Al Viro6d125522009-12-24 06:58:56 -05001785 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Trond Myklebust834f2a42005-10-18 14:20:16 -07001787 /* O_TRUNC implies we need access checks for write permissions */
Al Viro4296e2c2009-12-24 07:15:41 -05001788 if (open_flag & O_TRUNC)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001789 acc_mode |= MAY_WRITE;
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 /* Allow the LSM permission hook to distinguish append
1792 access from general write access. */
Al Viro4296e2c2009-12-24 07:15:41 -05001793 if (open_flag & O_APPEND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 acc_mode |= MAY_APPEND;
1795
Al Viro1f36f772009-12-26 10:56:19 -05001796 /* find the parent */
Al Viro9850c052010-01-13 15:01:15 -05001797reval:
Al Viro9b4a9b12009-04-07 11:44:16 -04001798 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001800 return ERR_PTR(error);
Al Viro9850c052010-01-13 15:01:15 -05001801 if (force_reval)
1802 nd.flags |= LOOKUP_REVAL;
Al Viro38662482009-12-26 07:21:48 -05001803
1804 current->total_link_count = 0;
1805 error = link_path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001806 if (error) {
Al Viro10fa8e62009-12-26 07:09:49 -05001807 filp = ERR_PTR(error);
1808 goto out;
J. R. Okajima654f5622009-06-18 23:30:15 +09001809 }
Al Viro1f36f772009-12-26 10:56:19 -05001810 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
Al Viro9b4a9b12009-04-07 11:44:16 -04001811 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
1813 /*
Al Viroa2c36b42009-12-24 03:39:50 -05001814 * We have the parent and last component.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
Al Viro8737f3a2008-08-02 22:36:57 -04001817 error = -ENFILE;
1818 filp = get_empty_filp();
1819 if (filp == NULL)
1820 goto exit_parent;
1821 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001822 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001823 nd.intent.open.flags = flag;
1824 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001825 nd.flags &= ~LOOKUP_PARENT;
Al Viro1f36f772009-12-26 10:56:19 -05001826 nd.flags |= LOOKUP_OPEN;
1827 if (open_flag & O_CREAT) {
1828 nd.flags |= LOOKUP_CREATE;
1829 if (open_flag & O_EXCL)
1830 nd.flags |= LOOKUP_EXCL;
1831 }
Al Viro3e297b62010-03-26 12:40:13 -04001832 if (open_flag & O_DIRECTORY)
1833 nd.flags |= LOOKUP_DIRECTORY;
Jan Kara002baee2010-05-13 12:52:57 +02001834 if (!(open_flag & O_NOFOLLOW))
1835 nd.flags |= LOOKUP_FOLLOW;
Al Viro3e297b62010-03-26 12:40:13 -04001836 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
Al Viro806b6812009-12-26 07:16:40 -05001837 while (unlikely(!filp)) { /* trailing symlink */
Al Virodef4af32009-12-26 08:37:05 -05001838 struct path holder;
Al Viro1f36f772009-12-26 10:56:19 -05001839 struct inode *inode = path.dentry->d_inode;
Al Virodef4af32009-12-26 08:37:05 -05001840 void *cookie;
Al Viro806b6812009-12-26 07:16:40 -05001841 error = -ELOOP;
Al Viro1f36f772009-12-26 10:56:19 -05001842 /* S_ISDIR part is a temporary automount kludge */
Jan Kara002baee2010-05-13 12:52:57 +02001843 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
Al Viro1f36f772009-12-26 10:56:19 -05001844 goto exit_dput;
1845 if (count++ == 32)
Al Viro806b6812009-12-26 07:16:40 -05001846 goto exit_dput;
1847 /*
1848 * This is subtle. Instead of calling do_follow_link() we do
1849 * the thing by hands. The reason is that this way we have zero
1850 * link_count and path_walk() (called from ->follow_link)
1851 * honoring LOOKUP_PARENT. After that we have the parent and
1852 * last component, i.e. we are in the same situation as after
1853 * the first path_walk(). Well, almost - if the last component
1854 * is normal we get its copy stored in nd->last.name and we will
1855 * have to putname() it when we are done. Procfs-like symlinks
1856 * just set LAST_BIND.
1857 */
1858 nd.flags |= LOOKUP_PARENT;
1859 error = security_inode_follow_link(path.dentry, &nd);
1860 if (error)
1861 goto exit_dput;
Al Virodef4af32009-12-26 08:37:05 -05001862 error = __do_follow_link(&path, &nd, &cookie);
1863 if (unlikely(error)) {
Al Viro806b6812009-12-26 07:16:40 -05001864 /* nd.path had been dropped */
Al Virodef4af32009-12-26 08:37:05 -05001865 if (!IS_ERR(cookie) && inode->i_op->put_link)
1866 inode->i_op->put_link(path.dentry, &nd, cookie);
1867 path_put(&path);
Al Viro806b6812009-12-26 07:16:40 -05001868 release_open_intent(&nd);
1869 filp = ERR_PTR(error);
1870 goto out;
1871 }
Al Virodef4af32009-12-26 08:37:05 -05001872 holder = path;
Al Viro806b6812009-12-26 07:16:40 -05001873 nd.flags &= ~LOOKUP_PARENT;
Al Viro3e297b62010-03-26 12:40:13 -04001874 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
Al Virodef4af32009-12-26 08:37:05 -05001875 if (inode->i_op->put_link)
1876 inode->i_op->put_link(holder.dentry, &nd, cookie);
1877 path_put(&holder);
Al Viro806b6812009-12-26 07:16:40 -05001878 }
Al Viro10fa8e62009-12-26 07:09:49 -05001879out:
Al Viro2a737872009-04-07 11:49:53 -04001880 if (nd.root.mnt)
1881 path_put(&nd.root);
Al Viro10fa8e62009-12-26 07:09:49 -05001882 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1883 force_reval = 1;
1884 goto reval;
1885 }
1886 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Al Viro806b6812009-12-26 07:16:40 -05001888exit_dput:
1889 path_put_conditional(&path, &nd);
1890 if (!IS_ERR(nd.intent.open.file))
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001891 release_open_intent(&nd);
Al Viro806b6812009-12-26 07:16:40 -05001892exit_parent:
1893 path_put(&nd.path);
1894 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05001895 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896}
1897
1898/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001899 * filp_open - open file and return file pointer
1900 *
1901 * @filename: path to open
1902 * @flags: open flags as per the open(2) second argument
1903 * @mode: mode for the new file if O_CREAT is set, else ignored
1904 *
1905 * This is the helper to open a file from kernelspace if you really
1906 * have to. But in generally you should not do this, so please move
1907 * along, nothing to see here..
1908 */
1909struct file *filp_open(const char *filename, int flags, int mode)
1910{
Al Viro6e8341a2009-04-06 11:16:22 -04001911 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001912}
1913EXPORT_SYMBOL(filp_open);
1914
1915/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 * lookup_create - lookup a dentry, creating it if it doesn't exist
1917 * @nd: nameidata info
1918 * @is_dir: directory flag
1919 *
1920 * Simple function to lookup and return a dentry and create it
1921 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001922 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001923 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 */
1925struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1926{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001927 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Jan Blunck4ac91372008-02-14 19:34:32 -08001929 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001930 /*
1931 * Yucky last component or no last component at all?
1932 * (foo/., foo/.., /////)
1933 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 if (nd->last_type != LAST_NORM)
1935 goto fail;
1936 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001937 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001938 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001939
1940 /*
1941 * Do the final lookup.
1942 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001943 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (IS_ERR(dentry))
1945 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001946
Al Viroe9baf6e2008-05-15 04:49:12 -04001947 if (dentry->d_inode)
1948 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001949 /*
1950 * Special case - lookup gave negative, but... we had foo/bar/
1951 * From the vfs_mknod() POV we just have a negative dentry -
1952 * all is fine. Let's be bastards - you had / on the end, you've
1953 * been asking for (non-existent) directory. -ENOENT for you.
1954 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001955 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1956 dput(dentry);
1957 dentry = ERR_PTR(-ENOENT);
1958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001960eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001962 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963fail:
1964 return dentry;
1965}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001966EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1969{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001970 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
1972 if (error)
1973 return error;
1974
1975 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1976 return -EPERM;
1977
Al Viroacfa4382008-12-04 10:06:33 -05001978 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 return -EPERM;
1980
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001981 error = devcgroup_inode_mknod(mode, dev);
1982 if (error)
1983 return error;
1984
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 error = security_inode_mknod(dir, dentry, mode, dev);
1986 if (error)
1987 return error;
1988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001990 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001991 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return error;
1993}
1994
Dave Hansen463c3192008-02-15 14:37:57 -08001995static int may_mknod(mode_t mode)
1996{
1997 switch (mode & S_IFMT) {
1998 case S_IFREG:
1999 case S_IFCHR:
2000 case S_IFBLK:
2001 case S_IFIFO:
2002 case S_IFSOCK:
2003 case 0: /* zero mode translates to S_IFREG */
2004 return 0;
2005 case S_IFDIR:
2006 return -EPERM;
2007 default:
2008 return -EINVAL;
2009 }
2010}
2011
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002012SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2013 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
Al Viro2ad94ae2008-07-21 09:32:51 -04002015 int error;
2016 char *tmp;
2017 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 struct nameidata nd;
2019
2020 if (S_ISDIR(mode))
2021 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Al Viro2ad94ae2008-07-21 09:32:51 -04002023 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002025 return error;
2026
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002028 if (IS_ERR(dentry)) {
2029 error = PTR_ERR(dentry);
2030 goto out_unlock;
2031 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002032 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002033 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002034 error = may_mknod(mode);
2035 if (error)
2036 goto out_dput;
2037 error = mnt_want_write(nd.path.mnt);
2038 if (error)
2039 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002040 error = security_path_mknod(&nd.path, dentry, mode, dev);
2041 if (error)
2042 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002043 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002045 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 break;
2047 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002048 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 new_decode_dev(dev));
2050 break;
2051 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002052 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002055out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002056 mnt_drop_write(nd.path.mnt);
2057out_dput:
2058 dput(dentry);
2059out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002060 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002061 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 putname(tmp);
2063
2064 return error;
2065}
2066
Heiko Carstens3480b252009-01-14 14:14:16 +01002067SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002068{
2069 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2070}
2071
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2073{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002074 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
2076 if (error)
2077 return error;
2078
Al Viroacfa4382008-12-04 10:06:33 -05002079 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 return -EPERM;
2081
2082 mode &= (S_IRWXUGO|S_ISVTX);
2083 error = security_inode_mkdir(dir, dentry, mode);
2084 if (error)
2085 return error;
2086
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002088 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002089 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 return error;
2091}
2092
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002093SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094{
2095 int error = 0;
2096 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002097 struct dentry *dentry;
2098 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
Al Viro2ad94ae2008-07-21 09:32:51 -04002100 error = user_path_parent(dfd, pathname, &nd, &tmp);
2101 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002102 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
Dave Hansen6902d922006-09-30 23:29:01 -07002104 dentry = lookup_create(&nd, 1);
2105 error = PTR_ERR(dentry);
2106 if (IS_ERR(dentry))
2107 goto out_unlock;
2108
Jan Blunck4ac91372008-02-14 19:34:32 -08002109 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002110 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002111 error = mnt_want_write(nd.path.mnt);
2112 if (error)
2113 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002114 error = security_path_mkdir(&nd.path, dentry, mode);
2115 if (error)
2116 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002117 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002118out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002119 mnt_drop_write(nd.path.mnt);
2120out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002121 dput(dentry);
2122out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002123 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002124 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002125 putname(tmp);
2126out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 return error;
2128}
2129
Heiko Carstens3cdad422009-01-14 14:14:22 +01002130SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002131{
2132 return sys_mkdirat(AT_FDCWD, pathname, mode);
2133}
2134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135/*
2136 * We try to drop the dentry early: we should have
2137 * a usage count of 2 if we're the only user of this
2138 * dentry, and if that is true (possibly after pruning
2139 * the dcache), then we drop the dentry now.
2140 *
2141 * A low-level filesystem can, if it choses, legally
2142 * do a
2143 *
2144 * if (!d_unhashed(dentry))
2145 * return -EBUSY;
2146 *
2147 * if it cannot handle the case of removing a directory
2148 * that is still in use by something else..
2149 */
2150void dentry_unhash(struct dentry *dentry)
2151{
2152 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002153 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 spin_lock(&dcache_lock);
2155 spin_lock(&dentry->d_lock);
2156 if (atomic_read(&dentry->d_count) == 2)
2157 __d_drop(dentry);
2158 spin_unlock(&dentry->d_lock);
2159 spin_unlock(&dcache_lock);
2160}
2161
2162int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2163{
2164 int error = may_delete(dir, dentry, 1);
2165
2166 if (error)
2167 return error;
2168
Al Viroacfa4382008-12-04 10:06:33 -05002169 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 return -EPERM;
2171
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002172 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 dentry_unhash(dentry);
2174 if (d_mountpoint(dentry))
2175 error = -EBUSY;
2176 else {
2177 error = security_inode_rmdir(dir, dentry);
2178 if (!error) {
2179 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002180 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002182 dont_mount(dentry);
2183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 }
2185 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002186 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 d_delete(dentry);
2189 }
2190 dput(dentry);
2191
2192 return error;
2193}
2194
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002195static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196{
2197 int error = 0;
2198 char * name;
2199 struct dentry *dentry;
2200 struct nameidata nd;
2201
Al Viro2ad94ae2008-07-21 09:32:51 -04002202 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002204 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002207 case LAST_DOTDOT:
2208 error = -ENOTEMPTY;
2209 goto exit1;
2210 case LAST_DOT:
2211 error = -EINVAL;
2212 goto exit1;
2213 case LAST_ROOT:
2214 error = -EBUSY;
2215 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002217
2218 nd.flags &= ~LOOKUP_PARENT;
2219
Jan Blunck4ac91372008-02-14 19:34:32 -08002220 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002221 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002223 if (IS_ERR(dentry))
2224 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002225 error = mnt_want_write(nd.path.mnt);
2226 if (error)
2227 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002228 error = security_path_rmdir(&nd.path, dentry);
2229 if (error)
2230 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002231 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002232exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002233 mnt_drop_write(nd.path.mnt);
2234exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002235 dput(dentry);
2236exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002237 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002239 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 putname(name);
2241 return error;
2242}
2243
Heiko Carstens3cdad422009-01-14 14:14:22 +01002244SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002245{
2246 return do_rmdir(AT_FDCWD, pathname);
2247}
2248
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249int vfs_unlink(struct inode *dir, struct dentry *dentry)
2250{
2251 int error = may_delete(dir, dentry, 0);
2252
2253 if (error)
2254 return error;
2255
Al Viroacfa4382008-12-04 10:06:33 -05002256 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 return -EPERM;
2258
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002259 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 if (d_mountpoint(dentry))
2261 error = -EBUSY;
2262 else {
2263 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002264 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002266 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002267 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002270 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
2272 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2273 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002274 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 }
Robert Love0eeca282005-07-12 17:06:03 -04002277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 return error;
2279}
2280
2281/*
2282 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002283 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 * writeout happening, and we don't want to prevent access to the directory
2285 * while waiting on the I/O.
2286 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002287static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288{
Al Viro2ad94ae2008-07-21 09:32:51 -04002289 int error;
2290 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 struct dentry *dentry;
2292 struct nameidata nd;
2293 struct inode *inode = NULL;
2294
Al Viro2ad94ae2008-07-21 09:32:51 -04002295 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002297 return error;
2298
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 error = -EISDIR;
2300 if (nd.last_type != LAST_NORM)
2301 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002302
2303 nd.flags &= ~LOOKUP_PARENT;
2304
Jan Blunck4ac91372008-02-14 19:34:32 -08002305 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002306 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 error = PTR_ERR(dentry);
2308 if (!IS_ERR(dentry)) {
2309 /* Why not before? Because we want correct error value */
2310 if (nd.last.name[nd.last.len])
2311 goto slashes;
2312 inode = dentry->d_inode;
2313 if (inode)
2314 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002315 error = mnt_want_write(nd.path.mnt);
2316 if (error)
2317 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002318 error = security_path_unlink(&nd.path, dentry);
2319 if (error)
2320 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002321 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002322exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002323 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 exit2:
2325 dput(dentry);
2326 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002327 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 if (inode)
2329 iput(inode); /* truncate the inode here */
2330exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002331 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 putname(name);
2333 return error;
2334
2335slashes:
2336 error = !dentry->d_inode ? -ENOENT :
2337 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2338 goto exit2;
2339}
2340
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002341SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002342{
2343 if ((flag & ~AT_REMOVEDIR) != 0)
2344 return -EINVAL;
2345
2346 if (flag & AT_REMOVEDIR)
2347 return do_rmdir(dfd, pathname);
2348
2349 return do_unlinkat(dfd, pathname);
2350}
2351
Heiko Carstens3480b252009-01-14 14:14:16 +01002352SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002353{
2354 return do_unlinkat(AT_FDCWD, pathname);
2355}
2356
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002357int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002359 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
2361 if (error)
2362 return error;
2363
Al Viroacfa4382008-12-04 10:06:33 -05002364 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 return -EPERM;
2366
2367 error = security_inode_symlink(dir, dentry, oldname);
2368 if (error)
2369 return error;
2370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002372 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002373 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 return error;
2375}
2376
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002377SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2378 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379{
Al Viro2ad94ae2008-07-21 09:32:51 -04002380 int error;
2381 char *from;
2382 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002383 struct dentry *dentry;
2384 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
2386 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002387 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002389
2390 error = user_path_parent(newdfd, newname, &nd, &to);
2391 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002392 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Dave Hansen6902d922006-09-30 23:29:01 -07002394 dentry = lookup_create(&nd, 0);
2395 error = PTR_ERR(dentry);
2396 if (IS_ERR(dentry))
2397 goto out_unlock;
2398
Dave Hansen75c3f292008-02-15 14:37:45 -08002399 error = mnt_want_write(nd.path.mnt);
2400 if (error)
2401 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002402 error = security_path_symlink(&nd.path, dentry, from);
2403 if (error)
2404 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002405 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002406out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002407 mnt_drop_write(nd.path.mnt);
2408out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002409 dput(dentry);
2410out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002411 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002412 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002413 putname(to);
2414out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 putname(from);
2416 return error;
2417}
2418
Heiko Carstens3480b252009-01-14 14:14:16 +01002419SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002420{
2421 return sys_symlinkat(oldname, AT_FDCWD, newname);
2422}
2423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2425{
2426 struct inode *inode = old_dentry->d_inode;
2427 int error;
2428
2429 if (!inode)
2430 return -ENOENT;
2431
Miklos Szeredia95164d2008-07-30 15:08:48 +02002432 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 if (error)
2434 return error;
2435
2436 if (dir->i_sb != inode->i_sb)
2437 return -EXDEV;
2438
2439 /*
2440 * A link to an append-only or immutable file cannot be created.
2441 */
2442 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2443 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002444 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002446 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 return -EPERM;
2448
2449 error = security_inode_link(old_dentry, dir, new_dentry);
2450 if (error)
2451 return error;
2452
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002453 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002455 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002456 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002457 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 return error;
2459}
2460
2461/*
2462 * Hardlinks are often used in delicate situations. We avoid
2463 * security-related surprises by not following symlinks on the
2464 * newname. --KAB
2465 *
2466 * We don't follow them on the oldname either to be compatible
2467 * with linux 2.0, and to avoid hard-linking to directories
2468 * and other special files. --ADM
2469 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002470SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2471 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472{
2473 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002474 struct nameidata nd;
2475 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002477 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002479 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002480 return -EINVAL;
2481
Al Viro2d8f3032008-07-22 09:59:21 -04002482 error = user_path_at(olddfd, oldname,
2483 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2484 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002486 return error;
2487
2488 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 if (error)
2490 goto out;
2491 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002492 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 goto out_release;
2494 new_dentry = lookup_create(&nd, 0);
2495 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002496 if (IS_ERR(new_dentry))
2497 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002498 error = mnt_want_write(nd.path.mnt);
2499 if (error)
2500 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002501 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2502 if (error)
2503 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002504 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002505out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002506 mnt_drop_write(nd.path.mnt);
2507out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002508 dput(new_dentry);
2509out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002510 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002512 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002513 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514out:
Al Viro2d8f3032008-07-22 09:59:21 -04002515 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
2517 return error;
2518}
2519
Heiko Carstens3480b252009-01-14 14:14:16 +01002520SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002521{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002522 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002523}
2524
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525/*
2526 * The worst of all namespace operations - renaming directory. "Perverted"
2527 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2528 * Problems:
2529 * a) we can get into loop creation. Check is done in is_subdir().
2530 * b) race potential - two innocent renames can create a loop together.
2531 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002532 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 * story.
2534 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002535 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 * whether the target exists). Solution: try to be smart with locking
2537 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002538 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 * move will be locked. Thus we can rank directories by the tree
2540 * (ancestors first) and rank all non-directories after them.
2541 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002542 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 * HOWEVER, it relies on the assumption that any object with ->lookup()
2544 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2545 * we'd better make sure that there's no link(2) for them.
2546 * d) some filesystems don't support opened-but-unlinked directories,
2547 * either because of layout or because they are not ready to deal with
2548 * all cases correctly. The latter will be fixed (taking this sort of
2549 * stuff into VFS), but the former is not going away. Solution: the same
2550 * trick as in rmdir().
2551 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002552 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002554 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 * locking].
2556 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002557static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2558 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559{
2560 int error = 0;
2561 struct inode *target;
2562
2563 /*
2564 * If we are going to change the parent - check write permissions,
2565 * we'll need to flip '..'.
2566 */
2567 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002568 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 if (error)
2570 return error;
2571 }
2572
2573 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2574 if (error)
2575 return error;
2576
2577 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04002578 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002579 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2581 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04002582 else {
2583 if (target)
2584 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04002588 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002590 dont_mount(new_dentry);
2591 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002592 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 if (d_unhashed(new_dentry))
2594 d_rehash(new_dentry);
2595 dput(new_dentry);
2596 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002597 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002598 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2599 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 return error;
2601}
2602
Adrian Bunk75c96f82005-05-05 16:16:09 -07002603static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2604 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605{
2606 struct inode *target;
2607 int error;
2608
2609 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2610 if (error)
2611 return error;
2612
2613 dget(new_dentry);
2614 target = new_dentry->d_inode;
2615 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002616 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2618 error = -EBUSY;
2619 else
2620 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2621 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05002622 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04002623 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07002624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 }
2627 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002628 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 dput(new_dentry);
2630 return error;
2631}
2632
2633int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2634 struct inode *new_dir, struct dentry *new_dentry)
2635{
2636 int error;
2637 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002638 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639
2640 if (old_dentry->d_inode == new_dentry->d_inode)
2641 return 0;
2642
2643 error = may_delete(old_dir, old_dentry, is_dir);
2644 if (error)
2645 return error;
2646
2647 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002648 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 else
2650 error = may_delete(new_dir, new_dentry, is_dir);
2651 if (error)
2652 return error;
2653
Al Viroacfa4382008-12-04 10:06:33 -05002654 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 return -EPERM;
2656
Robert Love0eeca282005-07-12 17:06:03 -04002657 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2658
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 if (is_dir)
2660 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2661 else
2662 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05002663 if (!error)
2664 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002665 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04002666 fsnotify_oldname_free(old_name);
2667
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 return error;
2669}
2670
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002671SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2672 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673{
Al Viro2ad94ae2008-07-21 09:32:51 -04002674 struct dentry *old_dir, *new_dir;
2675 struct dentry *old_dentry, *new_dentry;
2676 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002678 char *from;
2679 char *to;
2680 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Al Viro2ad94ae2008-07-21 09:32:51 -04002682 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 if (error)
2684 goto exit;
2685
Al Viro2ad94ae2008-07-21 09:32:51 -04002686 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 if (error)
2688 goto exit1;
2689
2690 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002691 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 goto exit2;
2693
Jan Blunck4ac91372008-02-14 19:34:32 -08002694 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 error = -EBUSY;
2696 if (oldnd.last_type != LAST_NORM)
2697 goto exit2;
2698
Jan Blunck4ac91372008-02-14 19:34:32 -08002699 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 if (newnd.last_type != LAST_NORM)
2701 goto exit2;
2702
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002703 oldnd.flags &= ~LOOKUP_PARENT;
2704 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002705 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 trap = lock_rename(new_dir, old_dir);
2708
Christoph Hellwig49705b72005-11-08 21:35:06 -08002709 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 error = PTR_ERR(old_dentry);
2711 if (IS_ERR(old_dentry))
2712 goto exit3;
2713 /* source must exist */
2714 error = -ENOENT;
2715 if (!old_dentry->d_inode)
2716 goto exit4;
2717 /* unless the source is a directory trailing slashes give -ENOTDIR */
2718 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2719 error = -ENOTDIR;
2720 if (oldnd.last.name[oldnd.last.len])
2721 goto exit4;
2722 if (newnd.last.name[newnd.last.len])
2723 goto exit4;
2724 }
2725 /* source should not be ancestor of target */
2726 error = -EINVAL;
2727 if (old_dentry == trap)
2728 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002729 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 error = PTR_ERR(new_dentry);
2731 if (IS_ERR(new_dentry))
2732 goto exit4;
2733 /* target should not be an ancestor of source */
2734 error = -ENOTEMPTY;
2735 if (new_dentry == trap)
2736 goto exit5;
2737
Dave Hansen9079b1e2008-02-15 14:37:49 -08002738 error = mnt_want_write(oldnd.path.mnt);
2739 if (error)
2740 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002741 error = security_path_rename(&oldnd.path, old_dentry,
2742 &newnd.path, new_dentry);
2743 if (error)
2744 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 error = vfs_rename(old_dir->d_inode, old_dentry,
2746 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002747exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002748 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749exit5:
2750 dput(new_dentry);
2751exit4:
2752 dput(old_dentry);
2753exit3:
2754 unlock_rename(new_dir, old_dir);
2755exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002756 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002757 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002759 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002761exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 return error;
2763}
2764
Heiko Carstensa26eab22009-01-14 14:14:17 +01002765SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002766{
2767 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2768}
2769
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2771{
2772 int len;
2773
2774 len = PTR_ERR(link);
2775 if (IS_ERR(link))
2776 goto out;
2777
2778 len = strlen(link);
2779 if (len > (unsigned) buflen)
2780 len = buflen;
2781 if (copy_to_user(buffer, link, len))
2782 len = -EFAULT;
2783out:
2784 return len;
2785}
2786
2787/*
2788 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2789 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2790 * using) it for any given inode is up to filesystem.
2791 */
2792int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2793{
2794 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002795 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002796 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002797
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002799 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002800 if (IS_ERR(cookie))
2801 return PTR_ERR(cookie);
2802
2803 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2804 if (dentry->d_inode->i_op->put_link)
2805 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2806 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807}
2808
2809int vfs_follow_link(struct nameidata *nd, const char *link)
2810{
2811 return __vfs_follow_link(nd, link);
2812}
2813
2814/* get the link contents into pagecache */
2815static char *page_getlink(struct dentry * dentry, struct page **ppage)
2816{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002817 char *kaddr;
2818 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002820 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002822 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002824 kaddr = kmap(page);
2825 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2826 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827}
2828
2829int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2830{
2831 struct page *page = NULL;
2832 char *s = page_getlink(dentry, &page);
2833 int res = vfs_readlink(dentry,buffer,buflen,s);
2834 if (page) {
2835 kunmap(page);
2836 page_cache_release(page);
2837 }
2838 return res;
2839}
2840
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002841void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002843 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002845 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846}
2847
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002848void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002850 struct page *page = cookie;
2851
2852 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 kunmap(page);
2854 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 }
2856}
2857
Nick Piggin54566b22009-01-04 12:00:53 -08002858/*
2859 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2860 */
2861int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862{
2863 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002864 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002865 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002866 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002868 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2869 if (nofs)
2870 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871
NeilBrown7e53cac2006-03-25 03:07:57 -08002872retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002873 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002874 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002876 goto fail;
2877
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 kaddr = kmap_atomic(page, KM_USER0);
2879 memcpy(kaddr, symname, len-1);
2880 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002881
2882 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2883 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 if (err < 0)
2885 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002886 if (err < len-1)
2887 goto retry;
2888
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 mark_inode_dirty(inode);
2890 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891fail:
2892 return err;
2893}
2894
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002895int page_symlink(struct inode *inode, const char *symname, int len)
2896{
2897 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002898 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002899}
2900
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002901const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 .readlink = generic_readlink,
2903 .follow_link = page_follow_link_light,
2904 .put_link = page_put_link,
2905};
2906
Al Viro2d8f3032008-07-22 09:59:21 -04002907EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908EXPORT_SYMBOL(follow_down);
2909EXPORT_SYMBOL(follow_up);
2910EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2911EXPORT_SYMBOL(getname);
2912EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913EXPORT_SYMBOL(lookup_one_len);
2914EXPORT_SYMBOL(page_follow_link_light);
2915EXPORT_SYMBOL(page_put_link);
2916EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002917EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918EXPORT_SYMBOL(page_symlink);
2919EXPORT_SYMBOL(page_symlink_inode_operations);
2920EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002921EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002922EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002923EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002924EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925EXPORT_SYMBOL(unlock_rename);
2926EXPORT_SYMBOL(vfs_create);
2927EXPORT_SYMBOL(vfs_follow_link);
2928EXPORT_SYMBOL(vfs_link);
2929EXPORT_SYMBOL(vfs_mkdir);
2930EXPORT_SYMBOL(vfs_mknod);
2931EXPORT_SYMBOL(generic_permission);
2932EXPORT_SYMBOL(vfs_readlink);
2933EXPORT_SYMBOL(vfs_rename);
2934EXPORT_SYMBOL(vfs_rmdir);
2935EXPORT_SYMBOL(vfs_symlink);
2936EXPORT_SYMBOL(vfs_unlink);
2937EXPORT_SYMBOL(dentry_unhash);
2938EXPORT_SYMBOL(generic_readlink);